diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp
index 000cffa..b5b331b 100644
--- a/src/core/CL/CLKernelLibrary.cpp
+++ b/src/core/CL/CLKernelLibrary.cpp
@@ -143,6 +143,7 @@
     { "copy_plane", "channel_extract.cl" },
     { "copy_planes_3p", "channel_combine.cl" },
     { "copy_to_keypoint", "fast_corners.cl" },
+    { "depthwise_convolution_3x3", "depthwise_convolution.cl" },
     { "derivative", "derivative.cl" },
     { "dilate", "dilate.cl" },
     { "direct_convolution1x1", "direct_convolution1x1.cl" },
@@ -348,6 +349,10 @@
 #include "./cl_kernels/depth_convert.clembed"
     },
     {
+        "depthwise_convolution.cl",
+#include "./cl_kernels/depthwise_convolution.clembed"
+    },
+    {
         "derivative.cl",
 #include "./cl_kernels/derivative.clembed"
     },
diff --git a/src/core/CL/cl_kernels/depthwise_convolution.cl b/src/core/CL/cl_kernels/depthwise_convolution.cl
new file mode 100644
index 0000000..cbcdbf2
--- /dev/null
+++ b/src/core/CL/cl_kernels/depthwise_convolution.cl
@@ -0,0 +1,189 @@
+/*
+ * 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 "helpers.h"
+
+#if CONV_STRIDE_X == 1
+#define convolution1x3 convolution1x3_stride_1
+#elif CONV_STRIDE_X == 2
+#define convolution1x3 convolution1x3_stride_2
+#elif CONV_STRIDE_X == 3
+#define convolution1x3 convolution1x3_stride_3
+#else /* CONV_STRIDE_X */
+#error "Stride not supported"
+#endif /* CONV_STRIDE_X */
+
+/** Compute a 1D horizontal convolution of size 3 and stride 1 for floating point type.
+ *
+ * @param[in] left_pixel   Pointer to the left pixel.
+ * @param[in] left_coeff   Weight of the left pixel
+ * @param[in] middle_coeff Weight of the middle pixel
+ * @param[in] right_coeff  Weight of the right pixel
+ *
+ * @return a float2 containing 2 convoluted values.
+ */
+inline float2 convolution1x3_stride_1(__global const uchar *left_pixel,
+                                      const float           left_coeff,
+                                      const float           middle_coeff,
+                                      const float           right_coeff)
+{
+    float4 temp = vload4(0, (__global float *)left_pixel);
+
+    float2 left   = CONVERT(temp.s01, float2);
+    float2 middle = CONVERT(temp.s12, float2);
+    float2 right  = CONVERT(temp.s23, float2);
+
+    return left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;
+}
+
+/** Compute a 1D horizontal convolution of size 3 and stride 2 for floating point type.
+ *
+ * @param[in] left_pixel   Pointer to the left pixel.
+ * @param[in] left_coeff   Weight of the left pixel
+ * @param[in] middle_coeff Weight of the middle pixel
+ * @param[in] right_coeff  Weight of the right pixel
+ *
+ * @return a float2 containing 2 convoluted values.
+ */
+inline float2 convolution1x3_stride_2(__global const uchar *left_pixel,
+                                      const float           left_coeff,
+                                      const float           middle_coeff,
+                                      const float           right_coeff)
+{
+    float4 temp0 = vload4(0, (__global float *)left_pixel);
+    float  temp1 = *((__global float *)(left_pixel + 4 * sizeof(float)));
+
+    float2 left   = CONVERT(temp0.s02, float2);
+    float2 middle = CONVERT(temp0.s13, float2);
+    float2 right  = CONVERT((float2)(temp0.s2, temp1), float2);
+
+    return left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;
+}
+
+/** Compute a 1D horizontal convolution of size 3 and stride 3 for floating point type.
+ *
+ * @param[in] left_pixel   Pointer to the left pixel.
+ * @param[in] left_coeff   Weight of the left pixel
+ * @param[in] middle_coeff Weight of the middle pixel
+ * @param[in] right_coeff  Weight of the right pixel
+ *
+ * @return a float2 containing 2 convoluted values.
+ */
+inline float2 convolution1x3_stride_3(__global const uchar *left_pixel,
+                                      const float           left_coeff,
+                                      const float           middle_coeff,
+                                      const float           right_coeff)
+{
+    float4 temp0 = vload4(0, (__global float *)left_pixel);
+    float2 temp1 = vload2(0, (__global float *)(left_pixel + 4 * sizeof(float)));
+
+    float2 left   = CONVERT(temp0.s03, float2);
+    float2 middle = CONVERT((float2)(temp0.s1, temp1.s0), float2);
+    float2 right  = CONVERT((float2)(temp0.s2, temp1.s1), float2);
+
+    return left * (float2)left_coeff + middle * (float2)middle_coeff + right * (float2)right_coeff;
+}
+
+/** Apply a 3x3 convolution matrix to a single channel F32 input image and return the result.
+ *
+ * Convolution matrix layout:
+ *
+ * [ mat0, mat1, mat2 ]\n
+ * [ mat3, mat4, mat5 ]\n
+ * [ mat6, mat7, mat8 ]\n
+ *
+ * @param[in] src  A pointer to source Image structure
+ * @param[in] mat0 Coefficient from the convolution matrix
+ * @param[in] mat1 Coefficient from the convolution matrix
+ * @param[in] mat2 Coefficient from the convolution matrix
+ * @param[in] mat3 Coefficient from the convolution matrix
+ * @param[in] mat4 Coefficient from the convolution matrix
+ * @param[in] mat5 Coefficient from the convolution matrix
+ * @param[in] mat6 Coefficient from the convolution matrix
+ * @param[in] mat0 Coefficient from the convolution matrix
+ * @param[in] mat7 Coefficient from the convolution matrix
+ * @param[in] mat8 Coefficient from the convolution matrix
+ *
+ * @return a float2 containing 2 convoluted values.
+ */
+inline float2 convolution3x3(
+    Image      *src,
+    const float mat0, const float mat1, const float mat2,
+    const float mat3, const float mat4, const float mat5,
+    const float mat6, const float mat7, const float mat8)
+{
+    float2 pixels;
+
+    pixels = convolution1x3(offset(src, 0, 0), mat0, mat1, mat2);
+    pixels += convolution1x3(offset(src, 0, 1), mat3, mat4, mat5);
+    pixels += convolution1x3(offset(src, 0, 2), mat6, mat7, mat8);
+
+    return pixels;
+}
+
+/** This function computes the horizontal integral of the image.
+  *
+  * @param[in] src_ptr                               Pointer to the source image. Supported data types: U8
+  * @param[in] src_stride_x                          Stride of the source image in X dimension (in bytes)
+  * @param[in] src_step_x                            src_stride_x * number of elements along X processed per workitem(in bytes)
+  * @param[in] src_stride_y                          Stride of the source image in Y dimension (in bytes)
+  * @param[in] src_step_y                            src_stride_y * number of elements along Y processed per workitem(in bytes)
+  * @param[in] src_offset_first_element_in_bytes     The offset of the first element in the source image
+  * @param[in] src_stride_z                          Stride of the source tensor in Z dimension (in bytes)
+  * @param[in] src_step_z                            src_stride_z * number of elements along Y processed per workitem(in bytes)
+  * @param[in] dst_ptr                               Pointer to the destination tensor. Supported data types: F16/F32
+  * @param[in] dst_stride_x                          Stride of the destination tensor in X dimension (in bytes)
+  * @param[in] dst_step_x                            dst_stride_x * number of elements along X processed per workitem(in bytes)
+  * @param[in] dst_stride_y                          Stride of the destination tensor in Y dimension (in bytes)
+  * @param[in] dst_step_y                            dst_stride_y * number of elements along Y processed per workitem(in bytes)
+  * @param[in] dst_stride_z                          Stride of the destination tensor in Z dimension (in bytes)
+  * @param[in] dst_step_z                            dst_stride_z * number of elements along Y processed per workitem(in bytes)
+  * @param[in] dst_offset_first_element_in_bytes     The offset of the first element in the destination tensor
+  * @param[in] weights_ptr                           Pointer to the weights tensor. Supported data types: F16/F32
+  * @param[in] weights_stride_x                      Stride of the weights tensor in X dimension (in bytes)
+  * @param[in] weights_step_x                        weights_stride_x * number of elements along X processed per workitem(in bytes)
+  * @param[in] weights_stride_y                      Stride of the weights tensor in Y dimension (in bytes)
+  * @param[in] weights_step_y                        weights_stride_y * number of elements along Y processed per workitem(in bytes)
+  * @param[in] weights_stride_z                      Stride of the weights tensor in Z dimension (in bytes)
+  * @param[in] weights_step_z                        weights_stride_z * number of elements along Y processed per workitem(in bytes)
+  * @param[in] weights_offset_first_element_in_bytes The offset of the first element in the weights tensor
+  */
+
+__kernel void depthwise_convolution_3x3(TENSOR3D_DECLARATION(src), TENSOR3D_DECLARATION(dst), TENSOR3D_DECLARATION(weights))
+{
+    Image    src     = CONVERT_TENSOR3D_TO_IMAGE_STRUCT(src);
+    Image    dst     = CONVERT_TENSOR3D_TO_IMAGE_STRUCT(dst);
+    Tensor3D weights = CONVERT_TO_TENSOR3D_STRUCT(weights);
+
+    uchar3 offset          = (uchar3)(0, 1, 2) * (uchar3)weights_stride_y;
+    float3 weights_values0 = vload3(0, (__global float *)(weights.ptr + offset.s0));
+    float3 weights_values1 = vload3(0, (__global float *)(weights.ptr + offset.s1));
+    float3 weights_values2 = vload3(0, (__global float *)(weights.ptr + offset.s2));
+
+    float2 pixels = convolution3x3(&src, weights_values0.s0, weights_values0.s1, weights_values0.s2,
+                                   weights_values1.s0, weights_values1.s1, weights_values1.s2,
+                                   weights_values2.s0, weights_values2.s1, weights_values2.s2);
+
+    vstore2(pixels, 0, (__global float *)dst.ptr);
+}
\ No newline at end of file
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionKernel.cpp
new file mode 100644
index 0000000..a24e304
--- /dev/null
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionKernel.cpp
@@ -0,0 +1,120 @@
+/*
+ * 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/CL/kernels/CLDepthwiseConvolutionKernel.h"
+
+#include "arm_compute/core/AccessWindowStatic.h"
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLKernel.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Utils.h"
+
+using namespace arm_compute;
+
+CLDepthwiseConvolutionKernel::CLDepthwiseConvolutionKernel()
+    : _border_size(0), _input(), _output(), _weights(), _conv_stride_x(0), _conv_stride_y(0), _conv_pad_x(0), _conv_pad_y(0)
+{
+}
+
+BorderSize CLDepthwiseConvolutionKernel::border_size() const
+{
+    return _border_size;
+}
+
+void CLDepthwiseConvolutionKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *weights, const PadStrideInfo &conv_info)
+{
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32);
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::F32);
+    ARM_COMPUTE_ERROR_ON(weights->info()->dimension(0) != 3 || weights->info()->dimension(1) != 3);
+
+    _input         = input;
+    _output        = output;
+    _weights       = weights;
+    _conv_stride_x = conv_info.stride().first;
+    _conv_stride_y = conv_info.stride().second;
+    _border_size   = BorderSize(weights->info()->dimension(1) / 2, weights->info()->dimension(0) / 2);
+    _conv_pad_x    = std::min(border_size().right, conv_info.pad().first);
+    _conv_pad_y    = std::min(border_size().bottom, conv_info.pad().second);
+
+    // Set build options
+    std::set<std::string> options;
+
+    options.emplace("-DCONV_STRIDE_X=" + support::cpp11::to_string(_conv_stride_x));
+
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("depthwise_convolution_3x3", options));
+
+    // Configure kernel window
+    const unsigned int num_elems_processed_per_iteration = 2;
+    const unsigned int num_elems_written_per_iteration   = 2;
+    const unsigned int num_elems_read_per_iteration      = (_conv_stride_x == 1) ? 4 : (_conv_stride_x == 2) ? 5 : 6;
+    const unsigned int num_rows_read_per_iteration       = 3;
+
+    Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));
+
+    const int access_right  = border_size().left + ceil_to_multiple(border_size().left + input->info()->dimension(0), num_elems_read_per_iteration);
+    const int access_bottom = border_size().bottom + ceil_to_multiple(border_size().bottom + input->info()->dimension(1), num_rows_read_per_iteration);
+
+    AccessWindowStatic     input_access(input->info(), -border_size().left, -border_size().bottom, access_right, access_bottom);
+    AccessWindowHorizontal output_access(output->info(), 0, num_elems_written_per_iteration);
+    AccessWindowStatic     weights_access(weights->info(), 0, 0, weights->info()->dimension(0), weights->info()->dimension(1));
+
+    update_window_and_padding(win, input_access, weights_access, output_access);
+
+    output_access.set_valid_region(win, ValidRegion(Coordinates(), output->info()->tensor_shape()));
+
+    ICLKernel::configure(win);
+}
+
+void CLDepthwiseConvolutionKernel::run(const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
+
+    Window slice_in      = window.first_slice_window_3D();
+    Window slice_out     = window.first_slice_window_3D();
+    Window slice_weights = window.first_slice_window_3D();
+
+    slice_in.adjust(Window::DimX, -_conv_pad_x, true);
+    slice_in.adjust(Window::DimY, -_conv_pad_y, true);
+    slice_in.set_dimension_step(Window::DimX, window.x().step() * _conv_stride_x);
+    slice_in.set_dimension_step(Window::DimY, window.y().step() * _conv_stride_y);
+    slice_weights.set_dimension_step(Window::DimX, 0);
+    slice_weights.set_dimension_step(Window::DimY, 0);
+
+    do
+    {
+        unsigned int idx = 0;
+        add_3D_tensor_argument(idx, _input, slice_in);
+        add_3D_tensor_argument(idx, _output, slice_out);
+        add_3D_tensor_argument(idx, _weights, slice_weights);
+
+        enqueue(queue, *this, slice_out);
+    }
+    while(window.slide_window_slice_3D(slice_out));
+}
\ No newline at end of file