diff --git a/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl b/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl
index cbe8266..d0cf032 100644
--- a/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl
+++ b/src/core/CL/cl_kernels/direct_convolution_1x1_3x3_5x5_quantized.cl
@@ -243,9 +243,7 @@
 
     pixels0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(pixels0, output_multiplier, output_shift, 8);
     pixels0 = pixels0 + output_offset;
-    pixels0 = max(pixels0, 0);
-    pixels0 = min(pixels0, 255);
 
-    vstore8(convert_uchar8(pixels0), 0, (__global uchar *)dst.ptr);
+    vstore8(convert_uchar8_sat(pixels0), 0, (__global uchar *)dst.ptr);
 }
 #endif // defined(DATA_TYPE) && defined(STRIDE_X) && defined(WEIGHTS_DEPTH)
diff --git a/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp b/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp
index 8ef7068..7596925 100644
--- a/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLDirectConvolutionLayer.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h"
 
+#include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/CL/kernels/CLDirectConvolutionLayerKernel.h"
 #include "arm_compute/core/PixelValue.h"
 #include "arm_compute/core/Utils.h"
@@ -45,7 +46,12 @@
     _direct_conv_kernel.configure(input, weights, biases, output, conv_info);
 
     // Configure border handler
-    _input_border_handler.configure(input, _direct_conv_kernel.border_size(), BorderMode::CONSTANT, PixelValue(0));
+    PixelValue &&zero_value(0.f);
+    if(is_data_type_quantized_asymmetric(input->info()->data_type()))
+    {
+        zero_value = PixelValue(static_cast<uint8_t>(input->info()->quantization_info().offset));
+    }
+    _input_border_handler.configure(input, _direct_conv_kernel.border_size(), BorderMode::CONSTANT, zero_value);
 }
 
 Error CLDirectConvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info)
diff --git a/tests/validation/CL/DirectConvolutionLayer.cpp b/tests/validation/CL/DirectConvolutionLayer.cpp
index 08d53d5..e6a196a 100644
--- a/tests/validation/CL/DirectConvolutionLayer.cpp
+++ b/tests/validation/CL/DirectConvolutionLayer.cpp
@@ -196,7 +196,7 @@
 TEST_SUITE(Quantized)
 TEST_SUITE(QASYMM8)
 FIXTURE_DATA_TEST_CASE(Run, CLDirectConvolutionLayerQuantizedFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(data, framework::dataset::make("DataType", DataType::QASYMM8)),
-                                                                                                                    framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255, 127) })))
+                                                                                                                    framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255, 10) })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_qasymm8);
diff --git a/tests/validation/CPP/ConvolutionLayer.cpp b/tests/validation/CPP/ConvolutionLayer.cpp
index a767912..3884a93 100644
--- a/tests/validation/CPP/ConvolutionLayer.cpp
+++ b/tests/validation/CPP/ConvolutionLayer.cpp
@@ -210,8 +210,7 @@
 
     acc = asymm_rounding_divide_by_pow2(asymm_int_mult(acc, output_multiplier), output_shift);
     acc += output_offset;
-    acc = std::max<int32_t>(acc, 0);
-    acc = std::min<int32_t>(acc, 255);
+    acc = clamp<int32_t>(acc, 0, 255);
 
     // Store the result
     *out_ptr = acc;
diff --git a/tests/validation/fixtures/DirectConvolutionLayerFixture.h b/tests/validation/fixtures/DirectConvolutionLayerFixture.h
index 1ec4d31..b78f13a 100644
--- a/tests/validation/fixtures/DirectConvolutionLayerFixture.h
+++ b/tests/validation/fixtures/DirectConvolutionLayerFixture.h
@@ -73,7 +73,7 @@
         {
             case DataType::QASYMM8:
             {
-                std::uniform_int_distribution<uint8_t> distribution(0, 10);
+                std::uniform_int_distribution<uint8_t> distribution(0, 50);
                 library->fill(tensor, distribution, i);
                 break;
             }
@@ -86,7 +86,7 @@
             }
             case DataType::S32:
             {
-                std::uniform_int_distribution<int32_t> distribution(-1000, 1000);
+                std::uniform_int_distribution<int32_t> distribution(-5, 5);
                 library->fill(tensor, distribution, i);
                 break;
             }