diff --git a/tests/validation/CPP/ActivationLayer.cpp b/tests/validation/CPP/ActivationLayer.cpp
new file mode 100644
index 0000000..fa393be
--- /dev/null
+++ b/tests/validation/CPP/ActivationLayer.cpp
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "ActivationLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/Helpers.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<T> activation_layer(const SimpleTensor<T> &src, ActivationLayerInfo info)
+{
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const T a(info.a());
+    const T b(info.b());
+
+    for(int i = 0; i < src.num_elements(); ++i)
+    {
+        T x = src[i];
+
+        switch(info.activation())
+        {
+            case ActivationLayerInfo::ActivationFunction::ABS:
+                dst[i] = std::abs(x);
+                break;
+            case ActivationLayerInfo::ActivationFunction::LINEAR:
+                dst[i] = a * x + b;
+                break;
+            case ActivationLayerInfo::ActivationFunction::LOGISTIC:
+                dst[i] = static_cast<T>(1) / (static_cast<T>(1) + std::exp(-x));
+                break;
+            case ActivationLayerInfo::ActivationFunction::RELU:
+                dst[i] = std::max<T>(static_cast<T>(0), x);
+                break;
+            case ActivationLayerInfo::ActivationFunction::BOUNDED_RELU:
+                dst[i] = std::min<T>(a, std::max(static_cast<T>(0), x));
+                break;
+            case ActivationLayerInfo::ActivationFunction::LEAKY_RELU:
+                dst[i] = (x > 0) ? x : a * x;
+                break;
+            case ActivationLayerInfo::ActivationFunction::SOFT_RELU:
+                dst[i] = std::log(static_cast<T>(1) + std::exp(x));
+                break;
+            case ActivationLayerInfo::ActivationFunction::SQRT:
+                dst[i] = std::sqrt(x);
+                break;
+            case ActivationLayerInfo::ActivationFunction::SQUARE:
+                dst[i] = x * x;
+                break;
+            case ActivationLayerInfo::ActivationFunction::TANH:
+                dst[i] = a * std::tanh(b * x);
+                break;
+            default:
+                ARM_COMPUTE_ERROR("Unsupported activation function");
+        }
+    }
+
+    return dst;
+}
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type>
+SimpleTensor<T> activation_layer(const SimpleTensor<T> &src, ActivationLayerInfo info)
+{
+    using namespace fixed_point_arithmetic;
+
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const int            fixed_point_position = src.fixed_point_position();
+    const fixed_point<T> a(info.a(), fixed_point_position);
+    const fixed_point<T> b(info.b(), fixed_point_position);
+    const fixed_point<T> const_0(0, fixed_point_position);
+    const fixed_point<T> const_1(1, fixed_point_position);
+
+    for(int i = 0; i < src.num_elements(); ++i)
+    {
+        fixed_point<T> x(src[i], fixed_point_position, true);
+
+        switch(info.activation())
+        {
+            case ActivationLayerInfo::ActivationFunction::ABS:
+                dst[i] = abs(x).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::LINEAR:
+                dst[i] = add(b, mul(a, x)).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::LOGISTIC:
+                dst[i] = (const_1 / (const_1 + exp(-x))).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::RELU:
+                dst[i] = max(const_0, x).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::BOUNDED_RELU:
+                dst[i] = min(a, max(const_0, x)).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::LEAKY_RELU:
+                dst[i] = (x > const_0) ? x.raw() : mul(a, x).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::SOFT_RELU:
+                dst[i] = log(const_1 + exp(x)).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::SQRT:
+                dst[i] = (const_1 / inv_sqrt(x)).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::SQUARE:
+                dst[i] = mul(x, x).raw();
+                break;
+            case ActivationLayerInfo::ActivationFunction::TANH:
+                dst[i] = mul(a, tanh(mul(b, x))).raw();
+                break;
+            default:
+                ARM_COMPUTE_ERROR("Unsupported activation function");
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> activation_layer(const SimpleTensor<float> &src, ActivationLayerInfo info);
+template SimpleTensor<half_float::half> activation_layer(const SimpleTensor<half_float::half> &src, ActivationLayerInfo info);
+template SimpleTensor<qint8_t> activation_layer(const SimpleTensor<qint8_t> &src, ActivationLayerInfo info);
+template SimpleTensor<qint16_t> activation_layer(const SimpleTensor<qint16_t> &src, ActivationLayerInfo info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/ActivationLayer.h b/tests/validation/CPP/ActivationLayer.h
new file mode 100644
index 0000000..09f602f
--- /dev/null
+++ b/tests/validation/CPP/ActivationLayer.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_ACTIVATION_LAYER_H__
+#define __ARM_COMPUTE_TEST_ACTIVATION_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+SimpleTensor<T> activation_layer(const SimpleTensor<T> &src, ActivationLayerInfo info);
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+SimpleTensor<T> activation_layer(const SimpleTensor<T> &src, ActivationLayerInfo info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_ACTIVATION_LAYER_H__ */
diff --git a/tests/validation/CPP/BitwiseAnd.cpp b/tests/validation/CPP/BitwiseAnd.cpp
new file mode 100644
index 0000000..6fc46b4
--- /dev/null
+++ b/tests/validation/CPP/BitwiseAnd.cpp
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "BitwiseAnd.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_and(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2)
+{
+    SimpleTensor<T> dst(src1.shape(), src1.data_type());
+
+    for(int i = 0; i < src1.num_elements(); ++i)
+    {
+        dst[i] = src1[i] & src2[i];
+    }
+
+    return dst;
+}
+
+template SimpleTensor<uint8_t> bitwise_and(const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/BitwiseAnd.h b/tests/validation/CPP/BitwiseAnd.h
new file mode 100644
index 0000000..eba2fd6
--- /dev/null
+++ b/tests/validation/CPP/BitwiseAnd.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_BITWISE_AND_H__
+#define __ARM_COMPUTE_TEST_BITWISE_AND_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_and(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_BITWISE_AND_H__ */
diff --git a/tests/validation/CPP/BitwiseNot.cpp b/tests/validation/CPP/BitwiseNot.cpp
new file mode 100644
index 0000000..5a6a13b
--- /dev/null
+++ b/tests/validation/CPP/BitwiseNot.cpp
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "BitwiseNot.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_not(const SimpleTensor<T> &src)
+{
+    SimpleTensor<T> dst(src.shape(), src.data_type());
+
+    for(int i = 0; i < src.num_elements(); ++i)
+    {
+        dst[i] = ~src[i];
+    }
+
+    return dst;
+}
+
+template SimpleTensor<uint8_t> bitwise_not(const SimpleTensor<uint8_t> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/BitwiseNot.h b/tests/validation/CPP/BitwiseNot.h
new file mode 100644
index 0000000..b4206f9
--- /dev/null
+++ b/tests/validation/CPP/BitwiseNot.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_BITWISE_NOT_H__
+#define __ARM_COMPUTE_TEST_BITWISE_NOT_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_not(const SimpleTensor<T> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_BITWISE_NOT_H__ */
diff --git a/tests/validation/CPP/BitwiseOr.cpp b/tests/validation/CPP/BitwiseOr.cpp
new file mode 100644
index 0000000..fc258d5
--- /dev/null
+++ b/tests/validation/CPP/BitwiseOr.cpp
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "BitwiseOr.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_or(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2)
+{
+    SimpleTensor<T> dst(src1.shape(), src1.data_type());
+
+    for(int i = 0; i < src1.num_elements(); ++i)
+    {
+        dst[i] = src1[i] | src2[i];
+    }
+
+    return dst;
+}
+
+template SimpleTensor<uint8_t> bitwise_or(const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/BitwiseOr.h b/tests/validation/CPP/BitwiseOr.h
new file mode 100644
index 0000000..39158cb
--- /dev/null
+++ b/tests/validation/CPP/BitwiseOr.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_BITWISE_OR_H__
+#define __ARM_COMPUTE_TEST_BITWISE_OR_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_or(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_BITWISE_OR_H__ */
diff --git a/tests/validation/CPP/BitwiseXor.cpp b/tests/validation/CPP/BitwiseXor.cpp
new file mode 100644
index 0000000..b8d275d
--- /dev/null
+++ b/tests/validation/CPP/BitwiseXor.cpp
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "BitwiseXor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_xor(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2)
+{
+    SimpleTensor<T> dst(src1.shape(), src1.data_type());
+
+    for(int i = 0; i < src1.num_elements(); ++i)
+    {
+        dst[i] = src1[i] ^ src2[i];
+    }
+
+    return dst;
+}
+
+template SimpleTensor<uint8_t> bitwise_xor(const SimpleTensor<uint8_t> &src1, const SimpleTensor<uint8_t> &src2);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/BitwiseXor.h b/tests/validation/CPP/BitwiseXor.h
new file mode 100644
index 0000000..3e7721e
--- /dev/null
+++ b/tests/validation/CPP/BitwiseXor.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_BITWISE_XOR_H__
+#define __ARM_COMPUTE_TEST_BITWISE_XOR_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> bitwise_xor(const SimpleTensor<T> &src1, const SimpleTensor<T> &src2);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_BITWISE_XOR_H__ */
diff --git a/tests/validation/CPP/ConvolutionLayer.cpp b/tests/validation/CPP/ConvolutionLayer.cpp
new file mode 100644
index 0000000..1824ada
--- /dev/null
+++ b/tests/validation/CPP/ConvolutionLayer.cpp
@@ -0,0 +1,205 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "ConvolutionLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/Helpers.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+namespace
+{
+inline bool is_valid_pixel(int i, int min, int max)
+{
+    return (i >= min && i < max);
+}
+
+// 3D convolution for floating point type
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+void convolution3d(const T *in, const T *weights, const T *bias, T *out, int xi, int yi, int width_in, int height_in, int depth_in, int width_weights, int height_weights, int fixed_point_position)
+{
+    ARM_COMPUTE_UNUSED(fixed_point_position);
+
+    const int half_width_weights  = width_weights / 2;
+    const int half_height_weights = height_weights / 2;
+
+    // Reset accumulator
+    T acc(0);
+
+    // Compute a 2D convolution for each IFM and accumulate the result
+    for(int ifm = 0; ifm < depth_in; ++ifm)
+    {
+        // Compute the offset for the input slice
+        const int offset_slice_in = xi + yi * width_in + ifm * width_in * height_in;
+
+        // Compute 2D convolution
+        for(int yk = -half_height_weights; yk <= half_height_weights; ++yk)
+        {
+            for(int xk = -half_width_weights; xk <= half_width_weights; ++xk)
+            {
+                // Check if the pixel is out-of-bound
+                if(is_valid_pixel(xi + xk, 0, width_in) && is_valid_pixel(yi + yk, 0, height_in))
+                {
+                    const int idx = xk + half_width_weights;
+                    const int idy = yk + half_height_weights;
+
+                    const T i_value = in[offset_slice_in + xk + yk * width_in];
+                    const T w_value = weights[idx + idy * width_weights + ifm * width_weights * height_weights];
+
+                    acc += i_value * w_value;
+                }
+            }
+        }
+    }
+
+    // Accumulate the bias and store the result
+    *out = acc + (*bias);
+}
+
+// 3D convolution for fixed point type
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+void convolution3d(const T *in, const T *weights, const T *bias, T *out, int xi, int yi, int width_in, int height_in, int depth_in, int width_weights, int height_weights,
+                   int fixed_point_position)
+{
+    const int half_width_weights  = width_weights / 2;
+    const int half_height_weights = height_weights / 2;
+
+    using namespace fixed_point_arithmetic;
+    using promoted_type = fixed_point_arithmetic::traits::promote_t<T>;
+
+    // Reset accumulator
+    fixed_point<promoted_type> acc(0, fixed_point_position);
+
+    // Compute a 2D convolution for each IFM and accumulate the result
+    for(int ifm = 0; ifm < depth_in; ++ifm)
+    {
+        // Compute the offset for the input slice
+        const int offset_slice_in = xi + yi * width_in + ifm * width_in * height_in;
+
+        // Compute 2D convolution
+        for(int yk = -half_height_weights; yk <= half_height_weights; ++yk)
+        {
+            for(int xk = -half_width_weights; xk <= half_width_weights; ++xk)
+            {
+                // Check if the pixel is out-of-bound
+                if(is_valid_pixel(xi + xk, 0, width_in) && is_valid_pixel(yi + yk, 0, height_in))
+                {
+                    const int idx = xk + half_width_weights;
+                    const int idy = yk + half_height_weights;
+
+                    const fixed_point<promoted_type> i_value(in[offset_slice_in + xk + yk * width_in], fixed_point_position, true);
+                    const fixed_point<promoted_type> w_value(weights[idx + idy * width_weights + ifm * width_weights * height_weights], fixed_point_position, true);
+                    const fixed_point<promoted_type> iw = i_value * w_value;
+                    acc                                 = iw + acc;
+                }
+            }
+        }
+    }
+
+    // Get the bias
+    const fixed_point<promoted_type> b(*bias, fixed_point_position, true);
+
+    // Accumulate the bias and covert back
+    acc = acc + b;
+    fixed_point<T> res(acc);
+    *out = res.raw();
+}
+} // namespace
+
+template <typename T>
+SimpleTensor<T> convolution_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &weights, const SimpleTensor<T> &bias, const TensorShape &output_shape, const PadStrideInfo &info)
+{
+    // Create reference
+    SimpleTensor<T> dst{ output_shape, src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const int width_in       = src.shape().x();
+    const int height_in      = src.shape().y();
+    const int depth_in       = src.shape().z();
+    const int width_out      = dst.shape().x();
+    const int height_out     = dst.shape().y();
+    const int depth_out      = dst.shape().z();
+    const int width_weights  = weights.shape().x();
+    const int height_weights = weights.shape().y();
+    const int depth_weights  = weights.shape().z();
+    const int pad_xi         = std::min(static_cast<int>(info.pad().first), width_weights / 2);
+    const int pad_yi         = std::min(static_cast<int>(info.pad().second), height_weights / 2);
+    const int start_xi       = width_weights / 2 - pad_xi;
+    const int start_yi       = height_weights / 2 - pad_yi;
+    const int end_xi         = width_in - start_xi;
+    const int end_yi         = height_in - start_yi;
+    const int stride_xi      = info.stride().first;
+    const int stride_yi      = info.stride().second;
+    const int num_batches    = src.shape().total_size() / (width_in * height_in * depth_in);
+
+    for(int r = 0; r < num_batches; ++r)
+    {
+        for(int yi = start_yi; yi < end_yi; yi += stride_yi)
+        {
+            for(int xi = start_xi; xi < end_xi; xi += stride_xi)
+            {
+                for(int ofm = 0; ofm < depth_out; ++ofm)
+                {
+                    // Compute input and output offsets
+                    const int offset_in  = r * width_in * height_in * depth_in;
+                    const int xo         = (xi - start_xi) / stride_xi;
+                    const int yo         = (yi - start_yi) / stride_yi;
+                    const int offset_out = xo + yo * width_out + ofm * width_out * height_out + r * width_out * height_out * depth_out;
+
+                    // Compute 3D convolution
+                    convolution3d(src.data() + offset_in,
+                                  weights.data() + ofm * width_weights * height_weights * depth_weights,
+                                  bias.data() + ofm,
+                                  dst.data() + offset_out,
+                                  xi, yi,
+                                  width_in, height_in, depth_in,
+                                  width_weights, height_weights,
+                                  src.fixed_point_position());
+                }
+            }
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> convolution_layer(const SimpleTensor<float> &src, const SimpleTensor<float> &weights, const SimpleTensor<float> &bias, const TensorShape &output_shape,
+                                               const PadStrideInfo &info);
+template SimpleTensor<half_float::half> convolution_layer(const SimpleTensor<half_float::half> &src, const SimpleTensor<half_float::half> &weights, const SimpleTensor<half_float::half> &bias,
+                                                          const TensorShape &output_shape, const PadStrideInfo &info);
+template SimpleTensor<qint8_t> convolution_layer(const SimpleTensor<qint8_t> &src, const SimpleTensor<qint8_t> &weights, const SimpleTensor<qint8_t> &bias, const TensorShape &output_shape,
+                                                 const PadStrideInfo &info);
+template SimpleTensor<qint16_t> convolution_layer(const SimpleTensor<qint16_t> &src, const SimpleTensor<qint16_t> &weights, const SimpleTensor<qint16_t> &bias, const TensorShape &output_shape,
+                                                  const PadStrideInfo &info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/ConvolutionLayer.h b/tests/validation/CPP/ConvolutionLayer.h
new file mode 100644
index 0000000..117e846
--- /dev/null
+++ b/tests/validation/CPP/ConvolutionLayer.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_CONVOLUTION_LAYER_H__
+#define __ARM_COMPUTE_TEST_CONVOLUTION_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> convolution_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &weights, const SimpleTensor<T> &bias, const TensorShape &output_shape, const PadStrideInfo &info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_CONVOLUTION_LAYER_H__ */
diff --git a/tests/validation/CPP/DepthConcatenateLayer.cpp b/tests/validation/CPP/DepthConcatenateLayer.cpp
new file mode 100644
index 0000000..139d26f
--- /dev/null
+++ b/tests/validation/CPP/DepthConcatenateLayer.cpp
@@ -0,0 +1,104 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "DepthConcatenateLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/Helpers.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> depthconcatenate_layer(const std::vector<SimpleTensor<T>> &srcs)
+{
+    // Create reference
+    std::vector<TensorShape> shapes;
+
+    for(const auto &src : srcs)
+    {
+        shapes.emplace_back(src.shape());
+    }
+
+    DataType        dst_type  = srcs.empty() ? DataType::UNKNOWN : srcs[0].data_type();
+    TensorShape     dst_shape = calculate_depth_concatenate_shape(shapes);
+    SimpleTensor<T> dst(dst_shape, dst_type);
+
+    // Compute reference
+    int       depth_offset = 0;
+    const int width_out    = dst.shape().x();
+    const int height_out   = dst.shape().y();
+    const int depth_out    = dst.shape().z();
+    const int out_stride_z = width_out * height_out;
+    const int batches      = dst.shape().total_size_upper(3);
+
+    // Set output tensor to 0
+    std::fill_n(dst.data(), dst.num_elements(), 0);
+
+    for(const auto &src : srcs)
+    {
+        ARM_COMPUTE_ERROR_ON(depth_offset >= depth_out);
+        ARM_COMPUTE_ERROR_ON(batches != static_cast<int>(src.shape().total_size_upper(3)));
+
+        const int width  = src.shape().x();
+        const int height = src.shape().y();
+        const int depth  = src.shape().z();
+        const int x_diff = (width_out - width) / 2;
+        const int y_diff = (height_out - height) / 2;
+
+        const T *src_ptr = src.data();
+
+        for(int b = 0; b < batches; ++b)
+        {
+            const size_t offset_to_first_element = b * out_stride_z * depth_out + depth_offset * out_stride_z + y_diff * width_out + x_diff;
+
+            for(int d = 0; d < depth; ++d)
+            {
+                for(int r = 0; r < height; ++r)
+                {
+                    std::copy(src_ptr, src_ptr + width, dst.data() + offset_to_first_element + d * out_stride_z + r * width_out);
+                    src_ptr += width;
+                }
+            }
+        }
+
+        depth_offset += depth;
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> depthconcatenate_layer(const std::vector<SimpleTensor<float>> &srcs);
+template SimpleTensor<half_float::half> depthconcatenate_layer(const std::vector<SimpleTensor<half_float::half>> &srcs);
+template SimpleTensor<qint8_t> depthconcatenate_layer(const std::vector<SimpleTensor<qint8_t>> &srcs);
+template SimpleTensor<qint16_t> depthconcatenate_layer(const std::vector<SimpleTensor<qint16_t>> &srcs);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/DepthConcatenateLayer.h b/tests/validation/CPP/DepthConcatenateLayer.h
new file mode 100644
index 0000000..3c486a8
--- /dev/null
+++ b/tests/validation/CPP/DepthConcatenateLayer.h
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_DEPTHCONCATENATE_LAYER_H__
+#define __ARM_COMPUTE_TEST_DEPTHCONCATENATE_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+
+#include <vector>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> depthconcatenate_layer(const std::vector<SimpleTensor<T>> &srcs);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_DEPTHCONCATENATE_LAYER_H__ */
diff --git a/tests/validation/CPP/DepthwiseConvolution.cpp b/tests/validation/CPP/DepthwiseConvolution.cpp
new file mode 100644
index 0000000..ebca333
--- /dev/null
+++ b/tests/validation/CPP/DepthwiseConvolution.cpp
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "DepthwiseConvolution.h"
+
+#include "ConvolutionLayer.h"
+#include "Utils.h"
+
+#include "tests/validation/Helpers.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+/** Perform a depthwise convolution
+ *
+ * - Three dimensions tensors
+ * - Third dimention is number of channels
+ * - Depths of input tensor and filter are equals
+ * - Padding, stride and output shape "match"
+ *
+ */
+template <typename T>
+SimpleTensor<T> depthwise_convolution(const SimpleTensor<T> &src, const SimpleTensor<T> &weights, const TensorShape &dst_shape, const PadStrideInfo &conv_info)
+{
+    // Create reference
+    SimpleTensor<T> dst{ dst_shape, src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const size_t filter_width  = weights.shape().x();
+    const size_t filter_height = weights.shape().y();
+    const size_t filter_plane  = filter_width * filter_height;
+    const size_t input_width   = src.shape().x();
+    const size_t input_height  = src.shape().y();
+    const size_t input_depth   = src.shape().z();
+
+    const size_t filter_half_size = filter_width / 2;
+    const size_t pad_x            = std::min(filter_half_size, static_cast<size_t>(conv_info.pad().first));
+    const size_t pad_y            = std::min(filter_half_size, static_cast<size_t>(conv_info.pad().second));
+    const size_t minimum_x        = -pad_x + filter_half_size;
+    const size_t minimum_y        = -pad_y + filter_half_size;
+
+    int out_pos = 0;
+    for(size_t z = 0; z < input_depth; ++z)
+    {
+        for(size_t y = minimum_y; y < input_height + pad_y - filter_half_size; y += conv_info.stride().second)
+        {
+            for(size_t x = minimum_x; x < input_width + pad_x - filter_half_size; x += conv_info.stride().first)
+            {
+                Coordinates coords(static_cast<int>(x), static_cast<int>(y), static_cast<int>(z));
+                size_t      filter_offset = filter_plane * z;
+
+                T val = 0;
+                for(int j = y - filter_half_size; j <= static_cast<int>(y + filter_half_size); ++j)
+                {
+                    for(int i = x - filter_half_size; i <= static_cast<int>(x + filter_half_size); ++i)
+                    {
+                        coords.set(0, i);
+                        coords.set(1, j);
+                        val += *(weights.data() + filter_offset) * tensor_elem_at(src, coords, BorderMode::CONSTANT, 0.f);
+                        ++filter_offset;
+                    }
+                }
+                coords.set(0, x);
+                coords.set(1, y);
+                dst[out_pos++] = saturate_cast<T>(val);
+            }
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> depthwise_convolution(const SimpleTensor<float> &src, const SimpleTensor<float> &weights, const TensorShape &dst_shape, const PadStrideInfo &conv_info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/DepthwiseConvolution.h b/tests/validation/CPP/DepthwiseConvolution.h
new file mode 100644
index 0000000..6be80fc
--- /dev/null
+++ b/tests/validation/CPP/DepthwiseConvolution.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_DEPTHWISE_CONVOLUTION_H__
+#define __ARM_COMPUTE_TEST_DEPTHWISE_CONVOLUTION_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> depthwise_convolution(const SimpleTensor<T> &src, const SimpleTensor<T> &weights, const TensorShape &dst_shape, const PadStrideInfo &conv_info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_DEPTHWISE_SEPARABLE_CONVOLUTION_LAYER_H__ */
diff --git a/tests/validation/CPP/DepthwiseSeparableConvolutionLayer.cpp b/tests/validation/CPP/DepthwiseSeparableConvolutionLayer.cpp
new file mode 100644
index 0000000..7020a85
--- /dev/null
+++ b/tests/validation/CPP/DepthwiseSeparableConvolutionLayer.cpp
@@ -0,0 +1,60 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "DepthwiseConvolution.h"
+
+#include "DepthwiseSeparableConvolutionLayer.h"
+
+#include "ConvolutionLayer.h"
+#include "Utils.h"
+
+#include "tests/validation/Helpers.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+// Depthwise separable convolution layer
+template <typename T>
+SimpleTensor<T> depthwise_separable_convolution_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &depthwise_weights, const TensorShape &depthwise_out_shape,
+                                                      const SimpleTensor<T> &pointwise_weights,
+                                                      const SimpleTensor<T> &biases, const TensorShape &dst_shape, const PadStrideInfo &depthwise_conv_info, const PadStrideInfo &pointwise_conv_info)
+{
+    // Compute reference
+    SimpleTensor<T> depthwise_out = depthwise_convolution(src, depthwise_weights, depthwise_out_shape, depthwise_conv_info);
+    SimpleTensor<T> dst           = convolution_layer(depthwise_out, pointwise_weights, biases, dst_shape, pointwise_conv_info);
+
+    return dst;
+}
+
+template SimpleTensor<float> depthwise_separable_convolution_layer(const SimpleTensor<float> &in, const SimpleTensor<float> &depthwise_weights, const TensorShape &depthwise_out_shape,
+                                                                   const SimpleTensor<float> &pointwise_weights, const SimpleTensor<float> &biases, const TensorShape &dst_shape, const PadStrideInfo &depthwise_conv_info, const PadStrideInfo &pointwise_conv_info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/DepthwiseSeparableConvolutionLayer.h b/tests/validation/CPP/DepthwiseSeparableConvolutionLayer.h
new file mode 100644
index 0000000..71cd013
--- /dev/null
+++ b/tests/validation/CPP/DepthwiseSeparableConvolutionLayer.h
@@ -0,0 +1,46 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_DEPTHWISE_SEPARABLE_CONVOLUTION_LAYER_H__
+#define __ARM_COMPUTE_TEST_DEPTHWISE_SEPARABLE_CONVOLUTION_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> depthwise_separable_convolution_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &depthwise_weights, const TensorShape &depthwise_out_shape,
+                                                      const SimpleTensor<T> &pointwise_weights,
+                                                      const SimpleTensor<T> &biases, const TensorShape &dst_shape, const PadStrideInfo &depthwise_conv_info, const PadStrideInfo &pointwise_conv_info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_DEPTHWISE_SEPARABLE_CONVOLUTION_LAYER_H__ */
diff --git a/tests/validation/CPP/DequantizationLayer.cpp b/tests/validation/CPP/DequantizationLayer.cpp
new file mode 100644
index 0000000..1c7ec25
--- /dev/null
+++ b/tests/validation/CPP/DequantizationLayer.cpp
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "DequantizationLayer.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type>
+SimpleTensor<float> dequantization_layer(const SimpleTensor<T> &src, float min, float max)
+{
+    // Create reference
+    SimpleTensor<float> dst{ src.shape(), DataType::F32 };
+
+    const float range   = max - min;
+    const float scaling = range / 255.0f;
+
+    for(int i = 0; i < src.num_elements(); ++i)
+    {
+        dst[i] = (static_cast<float>(src[i]) * scaling) + min;
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> dequantization_layer(const SimpleTensor<uint8_t> &src, float min, float max);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/DequantizationLayer.h b/tests/validation/CPP/DequantizationLayer.h
new file mode 100644
index 0000000..3aae338
--- /dev/null
+++ b/tests/validation/CPP/DequantizationLayer.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_DEQUANTIZATION_LAYER_H__
+#define __ARM_COMPUTE_TEST_DEQUANTIZATION_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+SimpleTensor<float> dequantization_layer(const SimpleTensor<T> &src, float min, float max);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_DEQUANTIZATION_LAYER_H__ */
diff --git a/tests/validation/CPP/Floor.cpp b/tests/validation/CPP/Floor.cpp
new file mode 100644
index 0000000..1c73944
--- /dev/null
+++ b/tests/validation/CPP/Floor.cpp
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "Floor.h"
+
+#include "tests/validation/Helpers.h"
+
+#include <cmath>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> floor_layer(const SimpleTensor<T> &src)
+{
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type() };
+
+    // Compute reference
+    for(int i = 0; i < src.num_elements(); ++i)
+    {
+        dst[i] = std::floor(src[i]);
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> floor_layer(const SimpleTensor<float> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/Floor.h b/tests/validation/CPP/Floor.h
new file mode 100644
index 0000000..d95ee30
--- /dev/null
+++ b/tests/validation/CPP/Floor.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_FLOOR_H__
+#define __ARM_COMPUTE_TEST_FLOOR_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> floor_layer(const SimpleTensor<T> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_FLOOR_H__ */
diff --git a/tests/validation/CPP/FullyConnectedLayer.cpp b/tests/validation/CPP/FullyConnectedLayer.cpp
new file mode 100644
index 0000000..a146535
--- /dev/null
+++ b/tests/validation/CPP/FullyConnectedLayer.cpp
@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "FullyConnectedLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/half.h"
+
+#include <numeric>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+namespace
+{
+// Vector matrix multiply for floating point
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+void vector_matrix_multiply(const T *src, const T *weights, const T *bias, T *dst, int cols_weights, int rows_weights, uint8_t fixed_point_position)
+{
+    ARM_COMPUTE_UNUSED(fixed_point_position);
+
+    for(int y = 0; y < rows_weights; ++y)
+    {
+        dst[y] = std::inner_product(src, src + cols_weights, weights, static_cast<T>(0)) + bias[y];
+        weights += cols_weights;
+    }
+}
+
+// Vector matrix multiply for fixed point type
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+void vector_matrix_multiply(const T *src, const T *weights, const T *bias, T *dst, int cols_weights, int rows_weights, uint8_t fixed_point_position)
+{
+    using namespace fixed_point_arithmetic;
+    using promoted_type = fixed_point_arithmetic::traits::promote_t<T>;
+
+    for(int y = 0; y < rows_weights; ++y)
+    {
+        // Reset accumulator
+        fixed_point<promoted_type> acc(0, fixed_point_position);
+
+        for(int x = 0; x < cols_weights; ++x)
+        {
+            const fixed_point<promoted_type> i_value(src[x], fixed_point_position, true);
+            const fixed_point<promoted_type> w_value(weights[x], fixed_point_position, true);
+            acc = acc + i_value * w_value;
+        }
+
+        // Get the bias
+        const fixed_point<T> b(bias[y], fixed_point_position, true);
+
+        // Convert back and accumulate the bias
+        fixed_point<T> res(acc);
+        res = res + b;
+
+        // Store the result
+        dst[y] = res.raw();
+
+        weights += cols_weights;
+    }
+}
+} // namespace
+
+template <typename T>
+SimpleTensor<T> fully_connected_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &weights, const SimpleTensor<T> &bias, const TensorShape &dst_shape)
+{
+    // Create reference
+    SimpleTensor<T> dst{ TensorShape{ dst_shape }, src.data_type(), 1, src.fixed_point_position() };
+
+    // Sanity checks
+    const int          num_batch_dimensions = std::max(0, static_cast<int>(dst_shape.num_dimensions()) - 1);
+    const int          num_input_dimensions = src.shape().num_dimensions() - num_batch_dimensions;
+    const unsigned int linear_input_size    = src.shape().total_size_lower(num_input_dimensions);
+
+    ARM_COMPUTE_UNUSED(num_batch_dimensions);
+    ARM_COMPUTE_UNUSED(num_input_dimensions);
+    ARM_COMPUTE_UNUSED(linear_input_size);
+    ARM_COMPUTE_ERROR_ON(weights.shape().x() != linear_input_size);
+    ARM_COMPUTE_ERROR_ON(weights.shape().y() != bias.shape().x());
+    ARM_COMPUTE_ERROR_ON(weights.shape().y() != dst.shape().x());
+
+    // Compute reference
+    const int cols_weights = weights.shape().x();
+    const int rows_weights = weights.shape().y();
+    const int num_batches  = dst_shape.total_size_upper(1);
+
+    for(int k = 0; k < num_batches; ++k)
+    {
+        vector_matrix_multiply<T>(src.data() + k * cols_weights,
+                                  weights.data(),
+                                  bias.data(),
+                                  dst.data() + k * rows_weights,
+                                  cols_weights,
+                                  rows_weights,
+                                  src.fixed_point_position());
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> fully_connected_layer(const SimpleTensor<float> &src, const SimpleTensor<float> &weights, const SimpleTensor<float> &bias, const TensorShape &dst_shape);
+template SimpleTensor<half_float::half> fully_connected_layer(const SimpleTensor<half_float::half> &src, const SimpleTensor<half_float::half> &weights, const SimpleTensor<half_float::half> &bias,
+                                                              const TensorShape &dst_shape);
+template SimpleTensor<qint8_t> fully_connected_layer(const SimpleTensor<qint8_t> &src, const SimpleTensor<qint8_t> &weights, const SimpleTensor<qint8_t> &bias, const TensorShape &dst_shape);
+template SimpleTensor<qint16_t> fully_connected_layer(const SimpleTensor<qint16_t> &src, const SimpleTensor<qint16_t> &weights, const SimpleTensor<qint16_t> &bias, const TensorShape &dst_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/FullyConnectedLayer.h b/tests/validation/CPP/FullyConnectedLayer.h
new file mode 100644
index 0000000..05c570a
--- /dev/null
+++ b/tests/validation/CPP/FullyConnectedLayer.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_FULLY_CONNECTED_LAYER_H__
+#define __ARM_COMPUTE_TEST_FULLY_CONNECTED_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> fully_connected_layer(const SimpleTensor<T> &src, const SimpleTensor<T> &weights, const SimpleTensor<T> &bias, const TensorShape &dst_shape);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_FULLY_CONNECTED_LAYER_H__ */
diff --git a/tests/validation/CPP/GEMM.cpp b/tests/validation/CPP/GEMM.cpp
new file mode 100644
index 0000000..9b66597
--- /dev/null
+++ b/tests/validation/CPP/GEMM.cpp
@@ -0,0 +1,122 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "GEMM.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<T> gemm(const SimpleTensor<T> &a, const SimpleTensor<T> &b, const SimpleTensor<T> &c, float alpha, float beta)
+{
+    // Create reference
+    SimpleTensor<T> dst{ c.shape(), c.data_type(), 1, c.fixed_point_position() };
+
+    // Compute reference
+    const int M = dst.shape().y();
+    const int N = dst.shape().x();
+    const int K = a.shape().x();
+
+    for(int row = 0; row < M; ++row)
+    {
+        for(int col = 0; col < N; ++col)
+        {
+            T acc(0);
+
+            for(int k = 0; k < K; ++k)
+            {
+                acc += a[row * K + k] * b[k * N + col];
+            }
+
+            // Finalize the result: alpha * A * B + beta * C
+            dst[col + row * N] = alpha * acc + beta * c[col + row * N];
+        }
+    }
+
+    return dst;
+}
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type>
+SimpleTensor<T> gemm(const SimpleTensor<T> &a, const SimpleTensor<T> &b, const SimpleTensor<T> &c, float alpha, float beta)
+{
+    using namespace fixed_point_arithmetic;
+
+    // Create reference
+    SimpleTensor<T> dst{ c.shape(), c.data_type(), 1, c.fixed_point_position() };
+
+    // Compute reference
+    using promoted_type = fixed_point_arithmetic::traits::promote_t<T>;
+
+    const int M                    = dst.shape().y();
+    const int N                    = dst.shape().x();
+    const int K                    = a.shape().x();
+    const int fixed_point_position = a.fixed_point_position();
+
+    const fixed_point<T> alpha_q(alpha, fixed_point_position);
+    const fixed_point<T> beta_q(beta, fixed_point_position);
+
+    for(int row = 0; row < M; ++row)
+    {
+        for(int col = 0; col < N; ++col)
+        {
+            fixed_point<promoted_type> acc_q(0, fixed_point_position);
+
+            for(int k = 0; k < K; ++k)
+            {
+                const fixed_point<promoted_type> a0_q(a[row * K + k], fixed_point_position, true);
+                const fixed_point<promoted_type> b0_q(b[k * N + col], fixed_point_position, true);
+
+                acc_q = acc_q + (a0_q * b0_q);
+            }
+
+            // Finalize the result: alpha * A * B + beta * C
+            const fixed_point<T> c0_q(c[col + row * N], fixed_point_position, true);
+
+            fixed_point<T> res_q(acc_q);
+            res_q = alpha_q * res_q;
+            res_q = res_q + (beta_q * c0_q);
+
+            // Store the result
+            dst[col + row * N] = res_q.raw();
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> gemm(const SimpleTensor<float> &a, const SimpleTensor<float> &b, const SimpleTensor<float> &c, float alpha, float beta);
+template SimpleTensor<half_float::half> gemm(const SimpleTensor<half_float::half> &a, const SimpleTensor<half_float::half> &b, const SimpleTensor<half_float::half> &c, float alpha, float beta);
+template SimpleTensor<qint8_t> gemm(const SimpleTensor<qint8_t> &a, const SimpleTensor<qint8_t> &b, const SimpleTensor<qint8_t> &c, float alpha, float beta);
+template SimpleTensor<qint16_t> gemm(const SimpleTensor<qint16_t> &a, const SimpleTensor<qint16_t> &b, const SimpleTensor<qint16_t> &c, float alpha, float beta);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/GEMM.h b/tests/validation/CPP/GEMM.h
new file mode 100644
index 0000000..cda792b
--- /dev/null
+++ b/tests/validation/CPP/GEMM.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_GEMM_H__
+#define __ARM_COMPUTE_TEST_GEMM_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+SimpleTensor<T> gemm(const SimpleTensor<T> &a, const SimpleTensor<T> &b, const SimpleTensor<T> &c, float alpha, float beta);
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+SimpleTensor<T> gemm(const SimpleTensor<T> &a, const SimpleTensor<T> &b, const SimpleTensor<T> &c, float alpha, float beta);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_GEMM_H__ */
diff --git a/tests/validation/CPP/L2Normalize.cpp b/tests/validation/CPP/L2Normalize.cpp
new file mode 100644
index 0000000..4fb4d57
--- /dev/null
+++ b/tests/validation/CPP/L2Normalize.cpp
@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "L2Normalize.h"
+#include "ReductionOperation.h"
+
+#include "tests/validation/Helpers.h"
+
+#include <algorithm>
+#include <cmath>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+namespace
+{
+TensorShape get_output_shape(TensorShape shape, unsigned int axis)
+{
+    TensorShape output_shape(shape);
+    output_shape.set(axis, 1);
+    return output_shape;
+}
+} // namespace
+
+template <typename T>
+SimpleTensor<T> l2_normalize(const SimpleTensor<T> &src, unsigned int axis, float epsilon)
+{
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type() };
+
+    // Reduce across given axis
+    SimpleTensor<T> sum = reduction_operation(src, get_output_shape(src.shape(), axis), axis, ReductionOperation::SUM_SQUARE);
+
+    // Compute reference
+    const int elems      = src.shape()[axis];
+    const int upper_dims = src.shape().total_size_upper(axis + 1);
+
+    for(int du = 0; du < upper_dims; ++du)
+    {
+        if(axis == 0)
+        {
+            const T *src_row_ptr         = src.data() + du * elems;
+            T       *dst_row_ptr         = dst.data() + du * elems;
+            const T  normalization_value = std::sqrt(std::max(sum[du], epsilon));
+            std::transform(src_row_ptr, src_row_ptr + elems, dst_row_ptr, [normalization_value](T val)
+            {
+                return val / normalization_value;
+            });
+        }
+        else
+        {
+            ARM_COMPUTE_ERROR("Unsupported normalization axis");
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> l2_normalize(const SimpleTensor<float> &src, unsigned int axis, float epsilon);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/L2Normalize.h b/tests/validation/CPP/L2Normalize.h
new file mode 100644
index 0000000..1db3ae6
--- /dev/null
+++ b/tests/validation/CPP/L2Normalize.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_L2NORMALIZE_H__
+#define __ARM_COMPUTE_TEST_L2NORMALIZE_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> l2_normalize(const SimpleTensor<T> &src, unsigned int axis, float epsilon);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_L2NORMALIZE_H__ */
diff --git a/tests/validation/CPP/MeanStdDev.cpp b/tests/validation/CPP/MeanStdDev.cpp
new file mode 100644
index 0000000..4a39b13
--- /dev/null
+++ b/tests/validation/CPP/MeanStdDev.cpp
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "MeanStdDev.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+std::pair<float, float> mean_and_standard_deviation(const SimpleTensor<T> &in)
+{
+    const int num_elements = in.num_elements();
+
+    // Calculate mean
+    float mean = std::accumulate(in.data(), in.data() + num_elements, 0.f) / num_elements;
+
+    // Calculate standard deviation
+    float std_dev = std::accumulate(in.data(), in.data() + num_elements, 0.f, [&mean](float a, float b)
+    {
+        return a + (mean - b) * (mean - b);
+    });
+
+    std_dev = std::sqrt(std_dev / num_elements);
+
+    return std::make_pair(mean, std_dev);
+}
+
+template std::pair<float, float> mean_and_standard_deviation(const SimpleTensor<uint8_t> &in);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/MeanStdDev.h b/tests/validation/CPP/MeanStdDev.h
new file mode 100644
index 0000000..6b89ae0
--- /dev/null
+++ b/tests/validation/CPP/MeanStdDev.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_STD_MEAN_DEV_H__
+#define __ARM_COMPUTE_TEST_STD_MEAN_DEV_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+std::pair<float, float> mean_and_standard_deviation(const SimpleTensor<T> &in);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_STD_MEAN_DEV_H__ */
diff --git a/tests/validation/CPP/NormalizationLayer.cpp b/tests/validation/CPP/NormalizationLayer.cpp
new file mode 100644
index 0000000..3c6f5e1
--- /dev/null
+++ b/tests/validation/CPP/NormalizationLayer.cpp
@@ -0,0 +1,275 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "NormalizationLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<T> normalization_layer(const SimpleTensor<T> &src, NormalizationLayerInfo info)
+{
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const uint32_t norm_size = info.norm_size();
+    NormType       type      = info.type();
+    float          beta      = info.beta();
+    uint32_t       kappa     = info.kappa();
+
+    const int cols       = src.shape()[0];
+    const int rows       = src.shape()[1];
+    const int depth      = src.shape()[2];
+    int       upper_dims = src.shape().total_size() / (cols * rows);
+
+    float coeff       = info.scale_coeff();
+    int   radius_cols = norm_size / 2;
+
+    // IN_MAP_1D and CROSS_MAP normalize over a single axis only
+    int radius_rows = (NormType::IN_MAP_2D == type) ? norm_size / 2 : 0;
+
+    if(type == NormType::CROSS_MAP)
+    {
+        // Remove also depth from upper dimensions since it is the dimension we
+        // want to use for normalization
+        upper_dims /= depth;
+
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int i = 0; i < rows; ++i)
+            {
+                for(int k = 0; k < cols; ++k)
+                {
+                    for(int l = 0; l < depth; ++l)
+                    {
+                        float accumulated_scale = 0.f;
+
+                        for(int j = -radius_cols; j <= radius_cols; ++j)
+                        {
+                            const int z = l + j;
+
+                            if(z >= 0 && z < depth)
+                            {
+                                const T value = src[k + i * cols + z * rows * cols + r * cols * rows * depth];
+                                accumulated_scale += value * value;
+                            }
+                        }
+
+                        dst[k + i * cols + l * rows * cols + r * cols * rows * depth] = kappa + accumulated_scale * coeff;
+                    }
+                }
+            }
+        }
+    }
+    else
+    {
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int i = 0; i < rows; ++i)
+            {
+                for(int k = 0; k < cols; ++k)
+                {
+                    float accumulated_scale = 0.f;
+
+                    for(int j = -radius_rows; j <= radius_rows; ++j)
+                    {
+                        const int y = i + j;
+                        for(int l = -radius_cols; l <= radius_cols; ++l)
+                        {
+                            const int x = k + l;
+
+                            if((x >= 0 && y >= 0) && (x < cols && y < rows))
+                            {
+                                const T value = src[x + y * cols + r * cols * rows];
+                                accumulated_scale += value * value;
+                            }
+                        }
+                    }
+
+                    dst[k + i * cols + r * cols * rows] = kappa + accumulated_scale * coeff;
+                }
+            }
+        }
+    }
+
+    if(beta == 1.f)
+    {
+        for(int i = 0; i < dst.num_elements(); ++i)
+        {
+            dst[i] = src[i] / dst[i];
+        }
+    }
+    else if(beta == 0.5f)
+    {
+        for(int i = 0; i < dst.num_elements(); ++i)
+        {
+            dst[i] = src[i] / std::sqrt(dst[i]);
+        }
+    }
+    else
+    {
+        for(int i = 0; i < dst.num_elements(); ++i)
+        {
+            dst[i] = src[i] * std::exp(std::log(dst[i]) * -beta);
+        }
+    }
+
+    return dst;
+}
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type>
+SimpleTensor<T> normalization_layer(const SimpleTensor<T> &src, NormalizationLayerInfo info)
+{
+    using namespace fixed_point_arithmetic;
+
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const int fixed_point_position = src.fixed_point_position();
+
+    const uint32_t norm_size = info.norm_size();
+    NormType       type      = info.type();
+    fixed_point<T> beta(info.beta(), fixed_point_position);
+    fixed_point<T> kappa(info.kappa(), fixed_point_position);
+
+    const int cols       = src.shape()[0];
+    const int rows       = src.shape()[1];
+    const int depth      = src.shape()[2];
+    int       upper_dims = src.shape().total_size() / (cols * rows);
+
+    fixed_point<T> coeff(info.scale_coeff(), fixed_point_position);
+    int            radius_cols = norm_size / 2;
+
+    // IN_MAP_1D and CROSS_MAP normalize over a single axis only
+    int radius_rows = (NormType::IN_MAP_2D == type) ? norm_size / 2 : 0;
+
+    if(type == NormType::CROSS_MAP)
+    {
+        // Remove also depth from upper dimensions since it is the dimension we
+        // want to use for normalization
+        upper_dims /= depth;
+
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int i = 0; i < rows; ++i)
+            {
+                for(int k = 0; k < cols; ++k)
+                {
+                    for(int l = 0; l < depth; ++l)
+                    {
+                        fixed_point<T> accumulated_scale(0.f, fixed_point_position);
+
+                        for(int j = -radius_cols; j <= radius_cols; ++j)
+                        {
+                            const int z = l + j;
+
+                            if(z >= 0 && z < depth)
+                            {
+                                const T              value = src[k + i * cols + z * rows * cols + r * cols * rows * depth];
+                                const fixed_point<T> fp_value(value, fixed_point_position, true);
+                                accumulated_scale = add(accumulated_scale, mul(fp_value, fp_value));
+                            }
+                        }
+
+                        accumulated_scale                                             = add(kappa, mul(accumulated_scale, coeff));
+                        dst[k + i * cols + l * rows * cols + r * cols * rows * depth] = accumulated_scale.raw();
+                    }
+                }
+            }
+        }
+    }
+    else
+    {
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int i = 0; i < rows; ++i)
+            {
+                for(int k = 0; k < cols; ++k)
+                {
+                    fixed_point<T> accumulated_scale(0.f, fixed_point_position);
+
+                    for(int j = -radius_rows; j <= radius_rows; ++j)
+                    {
+                        const int y = i + j;
+
+                        for(int l = -radius_cols; l <= radius_cols; ++l)
+                        {
+                            const int x = k + l;
+
+                            if((x >= 0 && y >= 0) && (x < cols && y < rows))
+                            {
+                                const T              value = src[x + y * cols + r * cols * rows];
+                                const fixed_point<T> fp_value(value, fixed_point_position, true);
+                                accumulated_scale = add(accumulated_scale, mul(fp_value, fp_value));
+                            }
+                        }
+                    }
+
+                    accumulated_scale                   = add(kappa, mul(accumulated_scale, coeff));
+                    dst[k + i * cols + r * cols * rows] = accumulated_scale.raw();
+                }
+            }
+        }
+    }
+
+    if(info.beta() == 1.f)
+    {
+        for(int i = 0; i < dst.num_elements(); ++i)
+        {
+            fixed_point<T> res = div(fixed_point<T>(src[i], fixed_point_position, true), fixed_point<T>(dst[i], fixed_point_position, true));
+            dst[i]             = res.raw();
+        }
+    }
+    else
+    {
+        const fixed_point<T> beta(info.beta(), fixed_point_position);
+
+        for(int i = 0; i < dst.num_elements(); ++i)
+        {
+            fixed_point<T> res = pow(fixed_point<T>(dst[i], fixed_point_position, true), beta);
+            res                = div(fixed_point<T>(src[i], fixed_point_position, true), res);
+            dst[i]             = res.raw();
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> normalization_layer(const SimpleTensor<float> &src, NormalizationLayerInfo info);
+template SimpleTensor<half_float::half> normalization_layer(const SimpleTensor<half_float::half> &src, NormalizationLayerInfo info);
+template SimpleTensor<qint8_t> normalization_layer(const SimpleTensor<qint8_t> &src, NormalizationLayerInfo info);
+template SimpleTensor<qint16_t> normalization_layer(const SimpleTensor<qint16_t> &src, NormalizationLayerInfo info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/NormalizationLayer.h b/tests/validation/CPP/NormalizationLayer.h
new file mode 100644
index 0000000..3f624ff
--- /dev/null
+++ b/tests/validation/CPP/NormalizationLayer.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_NORMALIZATION_LAYER_H__
+#define __ARM_COMPUTE_TEST_NORMALIZATION_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+SimpleTensor<T> normalization_layer(const SimpleTensor<T> &src, NormalizationLayerInfo info);
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+SimpleTensor<T> normalization_layer(const SimpleTensor<T> &src, NormalizationLayerInfo info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_NORMALIZATION_LAYER_H__ */
diff --git a/tests/validation/CPP/PoolingLayer.cpp b/tests/validation/CPP/PoolingLayer.cpp
new file mode 100644
index 0000000..c4425ca
--- /dev/null
+++ b/tests/validation/CPP/PoolingLayer.cpp
@@ -0,0 +1,243 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "PoolingLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+namespace
+{
+TensorShape calculate_output_shape(TensorShape shape, PoolingLayerInfo info)
+{
+    TensorShape dst_shape = shape;
+    const std::pair<unsigned int, unsigned int> scaled_dims = arm_compute::scaled_dimensions(shape.x(),
+                                                                                             shape.y(),
+                                                                                             info.pool_size(),
+                                                                                             info.pool_size(),
+                                                                                             info.pad_stride_info());
+    dst_shape.set(0, scaled_dims.first);
+    dst_shape.set(1, scaled_dims.second);
+
+    return dst_shape;
+}
+} // namespace
+
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, PoolingLayerInfo info)
+{
+    const int   pool_size     = info.pool_size();
+    PoolingType type          = info.pool_type();
+    int         pool_stride_x = info.pad_stride_info().stride().first;
+    int         pool_stride_y = info.pad_stride_info().stride().second;
+    int         pad_x         = info.pad_stride_info().pad().first;
+    int         pad_y         = info.pad_stride_info().pad().second;
+
+    const auto w_src      = static_cast<int>(src.shape()[0]);
+    const auto h_src      = static_cast<int>(src.shape()[1]);
+    const int  upper_dims = src.shape().total_size() / (w_src * h_src);
+
+    // Create reference
+    SimpleTensor<T> dst{ calculate_output_shape(src.shape(), info), src.data_type(), 1, src.fixed_point_position() };
+
+    const auto w_dst = static_cast<int>(dst.shape()[0]);
+    const auto h_dst = static_cast<int>(dst.shape()[1]);
+
+    if(type == PoolingType::MAX)
+    {
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int h = 0; h < h_dst; ++h)
+            {
+                for(int w = 0; w < w_dst; ++w)
+                {
+                    int wstart = w * pool_stride_x - pad_x;
+                    int hstart = h * pool_stride_y - pad_y;
+                    int wend   = std::min(wstart + pool_size, w_src);
+                    int hend   = std::min(hstart + pool_size, h_src);
+                    wstart     = std::max(wstart, 0);
+                    hstart     = std::max(hstart, 0);
+
+                    T max_val = std::numeric_limits<T>::lowest();
+                    for(int y = hstart; y < hend; ++y)
+                    {
+                        for(int x = wstart; x < wend; ++x)
+                        {
+                            const T val = src[r * h_src * w_src + y * w_src + x];
+                            if(val > max_val)
+                            {
+                                max_val = val;
+                            }
+                        }
+                    }
+
+                    dst[r * h_dst * w_dst + h * w_dst + w] = max_val;
+                }
+            }
+        }
+    }
+    else // Average pooling
+    {
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int h = 0; h < h_dst; ++h)
+            {
+                for(int w = 0; w < w_dst; ++w)
+                {
+                    T   avg_val(0);
+                    int wstart = w * pool_stride_x - pad_x;
+                    int hstart = h * pool_stride_y - pad_y;
+                    int wend   = std::min(wstart + pool_size, w_src + pad_x);
+                    int hend   = std::min(hstart + pool_size, h_src + pad_y);
+                    int pool   = (hend - hstart) * (wend - wstart);
+                    wstart     = std::max(wstart, 0);
+                    hstart     = std::max(hstart, 0);
+                    wend       = std::min(wend, w_src);
+                    hend       = std::min(hend, h_src);
+
+                    for(int y = hstart; y < hend; ++y)
+                    {
+                        for(int x = wstart; x < wend; ++x)
+                        {
+                            avg_val += src[r * h_src * w_src + y * w_src + x];
+                        }
+                    }
+                    dst[r * h_dst * w_dst + h * w_dst + w] = avg_val / pool;
+                }
+            }
+        }
+    }
+
+    return dst;
+}
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type>
+SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, PoolingLayerInfo info)
+{
+    const int   pool_size     = info.pool_size();
+    PoolingType type          = info.pool_type();
+    int         pool_stride_x = info.pad_stride_info().stride().first;
+    int         pool_stride_y = info.pad_stride_info().stride().second;
+    int         pad_x         = info.pad_stride_info().pad().first;
+    int         pad_y         = info.pad_stride_info().pad().second;
+
+    const auto w_src      = static_cast<int>(src.shape()[0]);
+    const auto h_src      = static_cast<int>(src.shape()[1]);
+    const int  upper_dims = src.shape().total_size() / (w_src * h_src);
+
+    // Create reference
+    SimpleTensor<T> dst{ calculate_output_shape(src.shape(), info), src.data_type(), 1, src.fixed_point_position() };
+
+    const auto w_dst = static_cast<int>(dst.shape()[0]);
+    const auto h_dst = static_cast<int>(dst.shape()[1]);
+
+    if(type == PoolingType::MAX)
+    {
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int h = 0; h < h_dst; ++h)
+            {
+                for(int w = 0; w < w_dst; ++w)
+                {
+                    int wstart = w * pool_stride_x - pad_x;
+                    int hstart = h * pool_stride_y - pad_y;
+                    int wend   = std::min(wstart + pool_size, w_src);
+                    int hend   = std::min(hstart + pool_size, h_src);
+                    wstart     = std::max(wstart, 0);
+                    hstart     = std::max(hstart, 0);
+
+                    T max_val = std::numeric_limits<T>::lowest();
+                    for(int y = hstart; y < hend; ++y)
+                    {
+                        for(int x = wstart; x < wend; ++x)
+                        {
+                            const T val = src[r * h_src * w_src + y * w_src + x];
+                            if(val > max_val)
+                            {
+                                max_val = val;
+                            }
+                        }
+                    }
+
+                    dst[r * h_dst * w_dst + h * w_dst + w] = max_val;
+                }
+            }
+        }
+    }
+    else // Average pooling
+    {
+        for(int r = 0; r < upper_dims; ++r)
+        {
+            for(int h = 0; h < h_dst; ++h)
+            {
+                for(int w = 0; w < w_dst; ++w)
+                {
+                    int wstart = w * pool_stride_x - pad_x;
+                    int hstart = h * pool_stride_y - pad_y;
+                    int wend   = std::min(wstart + pool_size, w_src + pad_x);
+                    int hend   = std::min(hstart + pool_size, h_src + pad_y);
+                    int pool   = (hend - hstart) * (wend - wstart);
+                    wstart     = std::max(wstart, 0);
+                    hstart     = std::max(hstart, 0);
+                    wend       = std::min(wend, w_src);
+                    hend       = std::min(hend, h_src);
+
+                    using namespace fixed_point_arithmetic;
+
+                    const int            fixed_point_position = src.fixed_point_position();
+                    const fixed_point<T> invpool_fp(1.f / static_cast<float>(pool), fixed_point_position);
+                    fixed_point<T>       avg_val(0, fixed_point_position, true);
+
+                    for(int y = hstart; y < hend; ++y)
+                    {
+                        for(int x = wstart; x < wend; ++x)
+                        {
+                            const fixed_point<T> in_fp(src[r * h_src * w_src + y * w_src + x], fixed_point_position, true);
+                            avg_val = add(avg_val, in_fp);
+                        }
+                    }
+                    dst[r * h_dst * w_dst + h * w_dst + w] = mul(avg_val, invpool_fp).raw();
+                }
+            }
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> pooling_layer(const SimpleTensor<float> &src, PoolingLayerInfo info);
+template SimpleTensor<half_float::half> pooling_layer(const SimpleTensor<half_float::half> &src, PoolingLayerInfo info);
+template SimpleTensor<qint8_t> pooling_layer(const SimpleTensor<qint8_t> &src, PoolingLayerInfo info);
+template SimpleTensor<qint16_t> pooling_layer(const SimpleTensor<qint16_t> &src, PoolingLayerInfo info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/PoolingLayer.h b/tests/validation/CPP/PoolingLayer.h
new file mode 100644
index 0000000..334054a
--- /dev/null
+++ b/tests/validation/CPP/PoolingLayer.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_POOLING_LAYER_H__
+#define __ARM_COMPUTE_TEST_POOLING_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, PoolingLayerInfo info);
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+SimpleTensor<T> pooling_layer(const SimpleTensor<T> &src, PoolingLayerInfo info);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_POOLING_LAYER_H__ */
diff --git a/tests/validation/CPP/QuantizationLayer.cpp b/tests/validation/CPP/QuantizationLayer.cpp
new file mode 100644
index 0000000..d61e75a
--- /dev/null
+++ b/tests/validation/CPP/QuantizationLayer.cpp
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "QuantizationLayer.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+void compute_min_max(const SimpleTensor<float> &src, float *min, float *max)
+{
+    // Set min and max to first pixel
+    float tmp_min = src[0];
+    float tmp_max = src[0];
+
+    // Look for min and max values
+    for(int i = 1; i < src.num_elements(); ++i)
+    {
+        if(src[i] < tmp_min)
+        {
+            tmp_min = src[i];
+        }
+        if(src[i] > tmp_max)
+        {
+            tmp_max = src[i];
+        }
+    }
+
+    *min = tmp_min;
+    *max = tmp_max;
+}
+
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<uint8_t> quantization_layer(const SimpleTensor<T> &src)
+{
+    // Create reference
+    SimpleTensor<uint8_t> dst{ src.shape(), DataType::U8 };
+
+    // Compute min and max of the tensor using Min-Max layer
+    float min = 0.f;
+    float max = 0.f;
+
+    compute_min_max(src, &min, &max);
+
+    const float range = max - min;
+
+    for(int i = 0; i < src.num_elements(); ++i)
+    {
+        // map values to range [0.0, 1.0]
+        const float normalized = (src[i] - min) / range;
+        dst[i]                 = static_cast<uint8_t>(std::min(255.0f, normalized * 256.0f));
+    }
+
+    return dst;
+}
+
+template SimpleTensor<uint8_t> quantization_layer(const SimpleTensor<float> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/QuantizationLayer.h b/tests/validation/CPP/QuantizationLayer.h
new file mode 100644
index 0000000..7c5572c
--- /dev/null
+++ b/tests/validation/CPP/QuantizationLayer.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_QUANTIZATION_LAYER_H__
+#define __ARM_COMPUTE_TEST_QUANTIZATION_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+SimpleTensor<uint8_t> quantization_layer(const SimpleTensor<T> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_QUANTIZATION_LAYER_H__ */
diff --git a/tests/validation/CPP/ReductionOperation.cpp b/tests/validation/CPP/ReductionOperation.cpp
new file mode 100644
index 0000000..acfcc09
--- /dev/null
+++ b/tests/validation/CPP/ReductionOperation.cpp
@@ -0,0 +1,93 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "ReductionOperation.h"
+
+#include "tests/validation/Helpers.h"
+
+#include <algorithm>
+#include <cmath>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+namespace
+{
+template <typename T>
+struct square
+{
+    T operator()(const T &lhs, const T &rhs) const
+    {
+        return (lhs + rhs * rhs);
+    }
+};
+
+template <typename T>
+T reduce_operation(T *ptr, int reduce_elements, ReductionOperation op)
+{
+    switch(op)
+    {
+        case ReductionOperation::SUM_SQUARE:
+            return std::accumulate(ptr, ptr + reduce_elements, 0.f, square<T>());
+        default:
+            ARM_COMPUTE_ERROR("Unsupported reduction operation");
+    }
+}
+} // namespace
+
+template <typename T>
+SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op)
+{
+    // Create reference
+    SimpleTensor<T> dst{ dst_shape, src.data_type() };
+
+    // Compute reference
+    const int reduce_elems = src.shape()[axis];
+    const int upper_dims   = src.shape().total_size_upper(axis + 1);
+
+    for(int du = 0; du < upper_dims; ++du)
+    {
+        if(axis == 0)
+        {
+            const T *src_row_ptr = src.data() + du * reduce_elems;
+            dst[du]              = reduce_operation(src_row_ptr, reduce_elems, op);
+        }
+        else
+        {
+            ARM_COMPUTE_ERROR("Unsupported reduction axis");
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> reduction_operation(const SimpleTensor<float> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/ReductionOperation.h b/tests/validation/CPP/ReductionOperation.h
new file mode 100644
index 0000000..6da6436
--- /dev/null
+++ b/tests/validation/CPP/ReductionOperation.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_REDUCTION_OPERATION_H__
+#define __ARM_COMPUTE_TEST_REDUCTION_OPERATION_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> reduction_operation(const SimpleTensor<T> &src, const TensorShape &dst_shape, unsigned int axis, ReductionOperation op);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_FLOOR_H__ */
diff --git a/tests/validation/CPP/Scale.cpp b/tests/validation/CPP/Scale.cpp
new file mode 100644
index 0000000..a1119f3
--- /dev/null
+++ b/tests/validation/CPP/Scale.cpp
@@ -0,0 +1,166 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "arm_compute/core/Helpers.h"
+
+#include "Scale.h"
+#include "Utils.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> scale(const SimpleTensor<T> &in, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value)
+{
+    TensorShape shape_scaled(in.shape());
+    shape_scaled.set(0, in.shape()[0] * scale_x);
+    shape_scaled.set(1, in.shape()[1] * scale_y);
+    SimpleTensor<T> out(shape_scaled, in.data_type());
+
+    // Compute the ratio between source width/height and destination width/height
+    const auto wr = static_cast<float>(in.shape()[0]) / static_cast<float>(out.shape()[0]);
+    const auto hr = static_cast<float>(in.shape()[1]) / static_cast<float>(out.shape()[1]);
+
+    const auto width  = static_cast<int>(in.shape().x());
+    const auto height = static_cast<int>(in.shape().y());
+
+    // Area interpolation behaves as Nearest Neighbour in case of up-sampling
+    if(policy == InterpolationPolicy::AREA && wr <= 1.f && hr <= 1.f)
+    {
+        policy = InterpolationPolicy::NEAREST_NEIGHBOR;
+    }
+
+    for(int element_idx = 0, count = 0; element_idx < out.num_elements(); ++element_idx, ++count)
+    {
+        Coordinates id    = index2coord(out.shape(), element_idx);
+        int         idx   = id.x();
+        int         idy   = id.y();
+        float       x_src = (idx + 0.5f) * wr - 0.5f;
+        float       y_src = (idy + 0.5f) * hr - 0.5f;
+
+        switch(policy)
+        {
+            case InterpolationPolicy::NEAREST_NEIGHBOR:
+            {
+                //Calculate the source coords without -0.5f is equivalent to round the x_scr/y_src coords
+                x_src = (idx + 0.5f) * wr;
+                y_src = (idy + 0.5f) * hr;
+                id.set(0, x_src);
+                id.set(1, y_src);
+
+                // If coordinates in range of tensor's width or height
+                if(x_src >= -1 || y_src >= -1 || x_src <= width || y_src <= height)
+                {
+                    out[element_idx] = tensor_elem_at(in, id, border_mode, constant_border_value);
+                }
+                else
+                {
+                    if(border_mode == BorderMode::CONSTANT)
+                    {
+                        out[element_idx] = constant_border_value;
+                    }
+                    else if(border_mode == BorderMode::REPLICATE)
+                    {
+                        id.set(0, clamp(static_cast<int>(x_src), 0, width - 1));
+                        id.set(1, clamp(static_cast<int>(y_src), 0, height - 1));
+                        out[element_idx] = in[coord2index(in.shape(), id)];
+                    }
+                }
+                break;
+            }
+            case InterpolationPolicy::BILINEAR:
+            {
+                id.set(0, std::floor(x_src));
+                id.set(1, std::floor(y_src));
+                if(x_src >= -1 || y_src >= -1 || x_src <= width || y_src <= height)
+                {
+                    out[element_idx] = bilinear_policy(in, id, x_src, y_src, border_mode, constant_border_value);
+                }
+                else
+                {
+                    if(border_mode == BorderMode::CONSTANT)
+                    {
+                        out[element_idx] = constant_border_value;
+                    }
+                    else if(border_mode == BorderMode::REPLICATE)
+                    {
+                        id.set(0, clamp(static_cast<int>(x_src), 0, width - 1));
+                        id.set(1, clamp(static_cast<int>(y_src), 0, height - 1));
+                        out[element_idx] = in[coord2index(in.shape(), id)];
+                    }
+                }
+                break;
+            }
+            case InterpolationPolicy::AREA:
+            {
+                int       x_from = std::floor(idx * wr - 0.5f - x_src);
+                int       y_from = std::floor(idy * hr - 0.5f - y_src);
+                int       x_to   = std::ceil((idx + 1) * wr - 0.5f - x_src);
+                int       y_to   = std::ceil((idy + 1) * hr - 0.5f - y_src);
+                const int xi     = std::floor(x_src);
+                const int yi     = std::floor(y_src);
+
+                // Clamp position to borders
+                x_src = std::max(-1.f, std::min(x_src, static_cast<float>(width)));
+                y_src = std::max(-1.f, std::min(y_src, static_cast<float>(height)));
+
+                // Clamp bounding box offsets to borders
+                x_from = ((x_src + x_from) < -1) ? -1 : x_from;
+                y_from = ((y_src + y_from) < -1) ? -1 : y_from;
+                x_to   = ((x_src + x_to) > width) ? (width - x_src) : x_to;
+                y_to   = ((y_src + y_to) > height) ? (height - y_src) : y_to;
+                ARM_COMPUTE_ERROR_ON((x_to - x_from + 1) == 0 || (y_to - y_from + 1) == 0);
+
+                float sum = 0;
+                for(int j = yi + y_from, je = yi + y_to; j <= je; ++j)
+                {
+                    for(int i = xi + x_from, ie = xi + x_to; i <= ie; ++i)
+                    {
+                        id.set(0, static_cast<int>(i));
+                        id.set(1, static_cast<int>(j));
+                        sum += tensor_elem_at(in, id, border_mode, constant_border_value);
+                    }
+                }
+                out[element_idx] = sum / ((x_to - x_from + 1) * (y_to - y_from + 1));
+
+                break;
+            }
+            default:
+                ARM_COMPUTE_ERROR("Unsupported interpolation mode");
+        }
+    }
+
+    return out;
+}
+
+template SimpleTensor<uint8_t> scale(const SimpleTensor<uint8_t> &src, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
\ No newline at end of file
diff --git a/tests/validation/CPP/Scale.h b/tests/validation/CPP/Scale.h
new file mode 100644
index 0000000..b882915
--- /dev/null
+++ b/tests/validation/CPP/Scale.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_SCALE_H__
+#define __ARM_COMPUTE_TEST_SCALE_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> scale(const SimpleTensor<T> &in, float scale_x, float scale_y, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value = 0);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_SCALE_H__ */
diff --git a/tests/validation/CPP/SoftmaxLayer.cpp b/tests/validation/CPP/SoftmaxLayer.cpp
new file mode 100644
index 0000000..4fe87d0
--- /dev/null
+++ b/tests/validation/CPP/SoftmaxLayer.cpp
@@ -0,0 +1,122 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "SoftmaxLayer.h"
+
+#include "tests/validation/FixedPoint.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type>
+SimpleTensor<T> softmax_layer(const SimpleTensor<T> &src)
+{
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const int cols       = src.shape()[0];
+    const int upper_dims = src.num_elements() / cols;
+
+    for(int r = 0; r < upper_dims; ++r)
+    {
+        const T *src_row_ptr = src.data() + r * cols;
+        T       *dst_row_ptr = dst.data() + r * cols;
+
+        // Find max
+        const T max = *std::max_element(src_row_ptr, src_row_ptr + cols);
+
+        // Regularize
+        T sum(0.f);
+        std::transform(src_row_ptr, src_row_ptr + cols, dst_row_ptr, [&sum, max](T val)
+        {
+            const T res(std::exp(val - max));
+            sum += res;
+            return res;
+        });
+
+        // Normalize
+        std::transform(dst_row_ptr, dst_row_ptr + cols, dst_row_ptr, [sum](T val)
+        {
+            return val / sum;
+        });
+    }
+
+    return dst;
+}
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type>
+SimpleTensor<T> softmax_layer(const SimpleTensor<T> &src)
+{
+    using namespace fixed_point_arithmetic;
+
+    // Create reference
+    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1, src.fixed_point_position() };
+
+    // Compute reference
+    const int cols       = src.shape()[0];
+    const int upper_dims = src.num_elements() / cols;
+
+    for(int r = 0; r < upper_dims; ++r)
+    {
+        const T *src_row_ptr = src.data() + r * cols;
+        T       *dst_row_ptr = dst.data() + r * cols;
+
+        // Find max
+        const fixed_point<T> max(*std::max_element(src_row_ptr, src_row_ptr + cols), src.fixed_point_position(), true);
+
+        // Regularize
+        using promoted_type = fixed_point_arithmetic::traits::promote_t<T>;
+        fixed_point<promoted_type> sum(0, src.fixed_point_position(), true);
+        std::transform(src_row_ptr, src_row_ptr + cols, dst_row_ptr, [&](T val)
+        {
+            const fixed_point<T> res = exp(fixed_point<T>(val, src.fixed_point_position(), true) - max);
+            sum                      = add(sum, fixed_point<promoted_type>(res.raw(), src.fixed_point_position(), true));
+            return res.raw();
+        });
+
+        // Normalize
+        fixed_point<T> saturated_sum(sum);
+        std::transform(dst_row_ptr, dst_row_ptr + cols, dst_row_ptr, [&](T val)
+        {
+            return div(fixed_point<T>(val, src.fixed_point_position(), true), saturated_sum).raw();
+        });
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> softmax_layer(const SimpleTensor<float> &src);
+template SimpleTensor<half_float::half> softmax_layer(const SimpleTensor<half_float::half> &src);
+template SimpleTensor<qint8_t> softmax_layer(const SimpleTensor<qint8_t> &src);
+template SimpleTensor<qint16_t> softmax_layer(const SimpleTensor<qint16_t> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/SoftmaxLayer.h b/tests/validation/CPP/SoftmaxLayer.h
new file mode 100644
index 0000000..ab79bc4
--- /dev/null
+++ b/tests/validation/CPP/SoftmaxLayer.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_SOFTMAX_LAYER_H__
+#define __ARM_COMPUTE_TEST_SOFTMAX_LAYER_H__
+
+#include "tests/SimpleTensor.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T, typename std::enable_if<is_floating_point<T>::value, int>::type = 0>
+SimpleTensor<T> softmax_layer(const SimpleTensor<T> &src);
+
+template <typename T, typename std::enable_if<std::is_integral<T>::value, int>::type = 0>
+SimpleTensor<T> softmax_layer(const SimpleTensor<T> &src);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_SOFTMAX_LAYER_H__ */
diff --git a/tests/validation/CPP/Utils.cpp b/tests/validation/CPP/Utils.cpp
new file mode 100644
index 0000000..05443ea
--- /dev/null
+++ b/tests/validation/CPP/Utils.cpp
@@ -0,0 +1,93 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "Utils.h"
+
+#include "tests/validation/Helpers.h"
+#include "tests/validation/half.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+// Return a tensor element at a specified coordinate with different border modes
+template <typename T>
+T tensor_elem_at(const SimpleTensor<T> &in, Coordinates coord, BorderMode border_mode, T constant_border_value)
+{
+    const int  x      = coord.x();
+    const int  y      = coord.y();
+    const auto width  = static_cast<int>(in.shape().x());
+    const auto height = static_cast<int>(in.shape().y());
+
+    // If coordinates beyond range of tensor's width or height
+    if(x < 0 || y < 0 || x >= width || y >= height)
+    {
+        if(border_mode == BorderMode::REPLICATE)
+        {
+            coord.set(0, std::max(0, std::min(x, width - 1)));
+            coord.set(1, std::max(0, std::min(y, height - 1)));
+        }
+        else
+        {
+            return constant_border_value;
+        }
+    }
+    return in[coord2index(in.shape(), coord)];
+}
+template float tensor_elem_at(const SimpleTensor<float> &in, Coordinates coord, BorderMode border_mode, float constant_border_value);
+template uint8_t tensor_elem_at(const SimpleTensor<uint8_t> &in, Coordinates coord, BorderMode border_mode, uint8_t constant_border_value);
+
+// Return the bilinear value at a specified coordinate with different border modes
+template <typename T>
+T bilinear_policy(const SimpleTensor<T> &in, Coordinates id, float xn, float yn, BorderMode border_mode, uint8_t constant_border_value)
+{
+    int idx = std::floor(xn);
+    int idy = std::floor(yn);
+
+    const float dx   = xn - idx;
+    const float dy   = yn - idy;
+    const float dx_1 = 1.0f - dx;
+    const float dy_1 = 1.0f - dy;
+
+    id.set(0, idx);
+    id.set(1, idy);
+    const T tl = tensor_elem_at(in, id, border_mode, constant_border_value);
+    id.set(0, idx + 1);
+    id.set(1, idy);
+    const T tr = tensor_elem_at(in, id, border_mode, constant_border_value);
+    id.set(0, idx);
+    id.set(1, idy + 1);
+    const T bl = tensor_elem_at(in, id, border_mode, constant_border_value);
+    id.set(0, idx + 1);
+    id.set(1, idy + 1);
+    const T br = tensor_elem_at(in, id, border_mode, constant_border_value);
+
+    return tl * (dx_1 * dy_1) + tr * (dx * dy_1) + bl * (dx_1 * dy) + br * (dx * dy);
+}
+template uint8_t bilinear_policy(const SimpleTensor<uint8_t> &in, Coordinates id, float xn, float yn, BorderMode border_mode, uint8_t constant_border_value);
+
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/CPP/Utils.h b/tests/validation/CPP/Utils.h
new file mode 100644
index 0000000..5e9ec82
--- /dev/null
+++ b/tests/validation/CPP/Utils.h
@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2017 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_TEST_VALIDATION_UTILS_H__
+#define __ARM_COMPUTE_TEST_VALIDATION_UTILS_H__
+
+#include "arm_compute/core/Types.h"
+#include "tests/Globals.h"
+#include "tests/ILutAccessor.h"
+#include "tests/Types.h"
+
+#include <array>
+#include <random>
+#include <type_traits>
+#include <utility>
+#include <vector>
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+template <typename T>
+T tensor_elem_at(const SimpleTensor<T> &in, Coordinates coord, BorderMode border_mode, T constant_border_value);
+
+template <typename T>
+T bilinear_policy(const SimpleTensor<T> &in, Coordinates id, float xn, float yn, BorderMode border_mode, uint8_t constant_border_value);
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_VALIDATION_UTILS_H__ */