COMPMID-617: Adds CLFullyConnectionLayer validation support

Change-Id: I4d2eb9872a3165fdcaa7784596e441cbe563dbc2
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/112577
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Ioan-Cristian Szabo <ioan-cristian.szabo@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

26 files changed, 1058 insertions, 302 deletions

diff --git a/arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h b/arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h
index c87fb2cd66..2520eff5de 100644
--- a/arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h
@@ -68,6 +68,14 @@ public:
      * @param[out] output Output tensor. Data type supported: same as @p input
      */
     void configure(const ICLTensor *input, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMInterleave4x4Kernel
+     *
+     * @param[in] input  Input tensor info. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/F16/U32/S32/F32
+     * @param[in] output Output tensor info which stores the interleaved matrix. Data type supported: same as @p input.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 
     // Inherited methods overridden
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h
index b60b80618c..3ad3ced003 100644
--- a/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h
@@ -61,6 +61,16 @@ public:
      * @param[in]  is_interleaved_transposed (Optional) True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel
      */
     void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, bool is_interleaved_transposed = true);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixMultiplyKernel
+     *
+     * @param[in] input0                    Input tensor info containing the interleaved Matrix A. Data type supported: QASYMM8
+     * @param[in] input1                    Input tensor info containing the transposed Matrix B. Data type supported: same as @p input0
+     * @param[in] output                    Output tensor info to store the result of matrix multiplication. Data type supported: S32
+     * @param[in] is_interleaved_transposed (Optional) True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, bool is_interleaved_transposed = true);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h
index 5f2e025687..871b97c1d7 100644
--- a/arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h
@@ -68,6 +68,19 @@ public:
      * @param[in]      b_offset       Offset to be added to each element of the matrix B.
      */
     void configure(ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, int32_t k, int32_t a_offset, int32_t b_offset);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpOffsetContributionKernel
+     *
+     * @param[in] mm_result      Input tensor containing the result of @ref CLGEMMLowpOffsetContributionKernel. Data type supported: S32
+     * @param[in] vector_sum_col Input row-vector of sums of all the entries in each column of matrix B.
+     *                           Note: vector_sum_col can be a nullptr in case a_offset = 0. Data type supported: same as @p mm_result
+     * @param[in] vector_sum_row Input row-vector of sums of all the entries in each row of matrix A.
+     *                           Note: vector_sum_row can be a nullptr in case b_offset = 0. Data type supported: same as @p mm_result
+     * @param[in] a_offset       Offset to be added to each element of the matrix A.
+     * @param[in] b_offset       Offset to be added to each element of the matrix B.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, int32_t a_offset, int32_t b_offset);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h
index aa0583fe81..12c12ef99a 100644
--- a/arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h
@@ -71,6 +71,14 @@ public:
      * @param[out] vector_sum_row Output row-vector of sums of all the entries in each row of mtx_a. Data type supported: S32
      */
     void configure(const ICLTensor *mtx_a, ICLTensor *vector_sum_row) override;
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixAReductionKernel
+     *
+     * @param[in] mtx_a          Input tensor. Data type supported: QASYMM8
+     * @param[in] vector_sum_row Output row-vector of sums of all the entries in each row of mtx_a. Data type supported: S32
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *mtx_a, const ITensorInfo *vector_sum_row);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
@@ -90,6 +98,14 @@ public:
      * @param[out] vector_sum_col Output row-vector of sums of all the entries in each column of mtx_b. Data type supported: S32
      */
     void configure(const ICLTensor *mtx_b, ICLTensor *vector_sum_col) override;
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixBReductionKernel
+     *
+     * @param[in] mtx_b          Input tensor. Data type supported: Data type supported: QASYMM8
+     * @param[in] vector_sum_col Output row-vector of sums of all the entries in each column of mtx_b. Data type supported: S32
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *mtx_b, const ITensorInfo *vector_sum_col);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.h b/arm_compute/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.h
index 9348ff8ca8..2956f93cdc 100644
--- a/arm_compute/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.h
@@ -50,6 +50,15 @@ public:
      * @param[in]      biases The shared biases tensor to append. It must be 1D tensor. Data types supported: Same as @p input
      */
     void configure(ICLTensor *accum, const ICLTensor *biases);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMMatrixAccumulateBiasesKernel
+     *
+     * @param[in] accum      The accumulate tensor to convert. Data types supported: QS8/QS16/F16/F32
+     * @param[in] biases     The shared biases tensor to append. It must be 1D tensor. Data types supported: Same as @p input
+     * @param[in] gpu_target GPU target
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *accum, const ITensorInfo *biases, GPUTarget gpu_target);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h b/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h
index 5af9091416..4e73d7eb13 100644
--- a/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h
@@ -60,6 +60,18 @@ public:
      * @param[in]  is_interleaved_transposed (Optional) True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel
      */
     void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, float alpha, bool is_interleaved_transposed = true);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMMatrixMultiplyKernel
+     *
+     * @param[in] input0                    Input tensor containing the Matrix A. Data types supported: QS8/QS16/F16/F32
+     * @param[in] input1                    Input tensor containing the Matrix B. Data type supported: same as @p input0
+     * @param[in] output                    Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0
+     * @param[in] alpha                     Weight of the matrix product
+     * @param[in] is_interleaved_transposed True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel
+     * @param[in] gpu_target                GPU Target
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved_transposed, GPUTarget gpu_target);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h b/arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h
index 8a37720462..8721643c1e 100644
--- a/arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMTranspose1xWKernel.h
@@ -74,6 +74,14 @@ public:
      * @param[out] output Output tensor. Data type supported: same as @p input
      */
     void configure(const ICLTensor *input, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMTranspose1xWKernel
+     *
+     * @param[in] input  Input tensor. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/F16/U32/S32/F32
+     * @param[in] output Output tensor. Data type supported: same as @p input.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/CL/kernels/CLIm2ColKernel.h b/arm_compute/core/CL/kernels/CLIm2ColKernel.h
index 1d8b5500c1..88de1ba002 100644
--- a/arm_compute/core/CL/kernels/CLIm2ColKernel.h
+++ b/arm_compute/core/CL/kernels/CLIm2ColKernel.h
@@ -69,7 +69,7 @@ public:
     /** Set the input and output of the kernel.
      *
      * @param[in]  input       The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
-     *                         while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QS16/QASYMM8/F16/F32
+     *                         while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QASYMM8/QS16/F16/F32
      * @param[out] output      The output tensor. First 2 lower dimensions represent a transform of each 3D input,
      *                         while every dimension above represents a batch. Data types supported: Same as @p input
      * @param[in]  kernel_dims The kernel dimensions (width and height).
@@ -80,6 +80,19 @@ public:
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
+    /** Static function to check if given info will lead to a valid configuration of @ref CLIm2ColKernel
+     *
+     * @param[in] input       The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
+     *                        while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QASYMM8/QS16/F16/F32
+     * @param[in] output      The output tensor. First 2 lower dimensions represent a transform of each 3D input,
+     *                        while every dimension above represents a batch. Data types supported: Same as @p input
+     * @param[in] kernel_dims The kernel dimensions (width and height).
+     * @param[in] conv_info   Contains padding and stride information described in @ref PadStrideInfo.
+     * @param[in] has_bias    In case biases are provided expands the matrix with 1.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias);
 
 private:
     /** Run the reshape kernel optimised for the special case (stride is 1, padding is 0 and kernel's low 3 dimensions are same as input)
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
new file mode 100644
index 0000000000..52773faa3a
--- /dev/null
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -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.
+ */
+#ifndef __ARM_COMPUTE_MISC_SHAPE_CALCULATOR_H__
+#define __ARM_COMPUTE_MISC_SHAPE_CALCULATOR_H__
+
+#include "arm_compute/core/ITensorInfo.h"
+
+namespace arm_compute
+{
+namespace misc
+{
+namespace shape_calculator
+{
+inline TensorShape compute_interleaved_shape(const ITensorInfo &a)
+{
+    // The interleaved output matrix will have the following shape: [ a_height * 4, ceil(a_width / 4.0f) ]
+    TensorShape shape_interleaved_a{ a.tensor_shape() };
+    shape_interleaved_a.set(0, a.dimension(0) * 4);
+    shape_interleaved_a.set(1, std::ceil(a.dimension(1) / 4.f));
+
+    return shape_interleaved_a;
+}
+inline TensorShape compute_transpose1xW_shape(const ITensorInfo &b)
+{
+    // The transpose1xW output matrix will have the following shape: [ b_height * 16, ceil(b_width / 16.0f) ]
+    TensorShape shape_transposed1xW_b{ b.tensor_shape() };
+    shape_transposed1xW_b.set(0, b.dimension(1) * 16);
+    shape_transposed1xW_b.set(1, std::ceil(b.dimension(0) / 16.f));
+
+    return shape_transposed1xW_b;
+}
+inline TensorShape compute_transpose1xW_with_element_size_shape(const ITensorInfo &b)
+{
+    // The transpose1xW output matrix will have the following shape:
+    // [ b_height * (16 / element_size), ceil(b_width / (16.0f / element_size) ]
+    TensorShape  shape_transposed1xW_b{ b.tensor_shape() };
+    const size_t transpose_width = 16 / b.element_size();
+    shape_transposed1xW_b.set(0, b.dimension(1) * transpose_width);
+    shape_transposed1xW_b.set(1, static_cast<size_t>(std::ceil(b.dimension(0) / static_cast<float>(transpose_width))));
+
+    return shape_transposed1xW_b;
+}
+inline TensorShape compute_reductionA_shape(const ITensorInfo &b)
+{
+    TensorShape shape_vector_sum_col{ b.tensor_shape() };
+    if(shape_vector_sum_col.num_dimensions() > 1)
+    {
+        shape_vector_sum_col.remove_dimension(1);
+    }
+
+    return shape_vector_sum_col;
+}
+inline TensorShape compute_reductionB_shape(const ITensorInfo &a)
+{
+    TensorShape shape_vector_sum_row{ a.tensor_shape() };
+    shape_vector_sum_row.set(Window::DimX, a.dimension(1));
+    if(a.num_dimensions() > 1)
+    {
+        shape_vector_sum_row.remove_dimension(1);
+    }
+
+    return shape_vector_sum_row;
+}
+inline TensorShape compute_im2col_shape(const ITensorInfo &input)
+{
+    TensorShape shape_im2col{ input.tensor_shape() };
+    shape_im2col.collapse(3);
+
+    return shape_im2col;
+}
+inline TensorShape compute_transposed_shape(const ITensorInfo &input)
+{
+    TensorShape shape_transposed{ input.tensor_shape() };
+
+    shape_transposed.set(0, input.dimension(1));
+    shape_transposed.set(1, input.dimension(0));
+
+    return shape_transposed;
+}
+} // namespace shape_calculator
+} // namespace misc
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_MISC_SHAPE_CALCULATOR_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
index 2cac06c1c9..1e9ee492ad 100644
--- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
@@ -52,6 +52,14 @@ public:
      * @param[out] output Destination tensor which stores the transposed input tensor. Data type supported: Same as @p input.
      */
     void configure(const ICLTensor *input, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLFullyConnectedLayerReshapeWeights
+     *
+     * @param[in] input  Weights tensor. The weights must be 2 dimensional. Data types supported: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in] output Destination tensor which stores the transposed input tensor. Data type supported: Same as @p input.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 };
 
 /** Basic function to compute a Fully Connected layer on OpenCL. This function calls the following OpenCL kernels:
@@ -78,6 +86,18 @@ public:
      * @param[in]  are_weights_reshaped (Optional) Reshape the weights tensor if false. Defaults to false.
      */
     void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, bool transpose_weights = true, bool are_weights_reshaped = false);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLFullyConnectedLayer
+     *
+     * @param[in] input                Source tensor. Data type supported: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in] weights              Weights tensor. The weights must be 2 dimensional. Data type supported: Same as @p input
+     * @param[in] biases               Bias tensor. It can be nullptr. Data type supported:Same as @p input.
+     * @param[in] output               Destination tensor. Data type supported: Same as @p input.
+     * @param[in] transpose_weights    (Optional) Transpose weights if true. Defaults to true.
+     * @param[in] are_weights_reshaped (Optional) Reshape the weights tensor if false. Defaults to false.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, bool transpose_weights = true, bool are_weights_reshaped = false);
 
     //Inherited methods override
     void run() override;
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
index e316144548..3976704907 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
@@ -69,6 +69,17 @@ public:
      *                       if the reshape of matrix B should be executed only for the first run
      */
     void configure(const ICLTensor *a, const ICLTensor *b, ICLTensor *output, const GEMMInfo &gemm_info = GEMMInfo());
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixMultiplyCore
+     *
+     * @param[in] a         First input tensor  (Matrix A). Data type supported: QASYMM8.
+     * @param[in] b         Second input tensor (Matrix B). Data type supported: same as @p a
+     * @param[in] output    Output tensor. Data type supported: Data type supported: S32
+     * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
+     *                      if the reshape of matrix B should be executed only for the first run
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *output, const GEMMInfo &gemm_info = GEMMInfo());
 
     // Inherited methods overridden:
     void run() override;
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
index 46e6b494f8..eddb3a26b7 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
@@ -75,11 +75,11 @@ public:
     void configure(const ITensor *a, const ITensor *b, ITensor *output, const GEMMInfo &gemm_info = GEMMInfo());
     /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixMultiplyCore
      *
-     * @param[in]  a         First input tensor  (Matrix A). Data type supported: QASYMM8.
-     * @param[in]  b         Second input tensor (Matrix B). Data type supported: same as @p a
-     * @param[out] output    Output tensor. Data type supported: Data type supported: S32
-     * @param[in]  gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
-     *                       if the reshape of matrix B should be executed only for the first run
+     * @param[in] a         First input tensor  (Matrix A). Data type supported: QASYMM8.
+     * @param[in] b         Second input tensor (Matrix B). Data type supported: same as @p a
+     * @param[in] output    Output tensor. Data type supported: Data type supported: S32
+     * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
+     *                      if the reshape of matrix B should be executed only for the first run
      *
      * @return a status
      */
diff --git a/src/core/CL/kernels/CLGEMMInterleave4x4Kernel.cpp b/src/core/CL/kernels/CLGEMMInterleave4x4Kernel.cpp
index 7741f12900..6886f54602 100644
--- a/src/core/CL/kernels/CLGEMMInterleave4x4Kernel.cpp
+++ b/src/core/CL/kernels/CLGEMMInterleave4x4Kernel.cpp
@@ -33,8 +33,55 @@
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
+
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::U8, DataType::S8,
+                                                         DataType::QS16, DataType::U16, DataType::S16, DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_interleaved_shape(*input));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    }
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
+{
+    unsigned int           num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(input->data_type());
+    constexpr unsigned int num_elems_processed_per_iteration_y = 4;
+    const unsigned int     num_elems_written_per_iteration     = num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y;
+    bool                   window_changed                      = false;
+
+    // Configure kernel window
+    Window                win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
+    AccessWindowRectangle input_access(input, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
+    window_changed = window_changed || update_window_and_padding(win, input_access);
+
+    // Configure window in case of configured output
+    if(output->total_size() != 0)
+    {
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_written_per_iteration, 1, 4.f, 0.25f);
+        window_changed = window_changed || update_window_and_padding(win, output_access);
+        output_access.set_valid_region(win, input->valid_region());
+    }
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
 
 CLGEMMInterleave4x4Kernel::CLGEMMInterleave4x4Kernel()
     : _input(nullptr), _output(nullptr)
@@ -43,22 +90,13 @@ CLGEMMInterleave4x4Kernel::CLGEMMInterleave4x4Kernel()
 
 void CLGEMMInterleave4x4Kernel::configure(const ICLTensor *input, ICLTensor *output)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::QASYMM8,
-                                                  DataType::U16, DataType::S16, DataType::QS16,
-                                                  DataType::U32, DataType::S32,
-                                                  DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_NULLPTR(output);
-
-    TensorShape output_shape = input->info()->tensor_shape();
-    output_shape.set(0, input->info()->dimension(0) * 4);
-    output_shape.set(1, std::ceil(input->info()->dimension(1) / 4.0f));
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
+    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(compute_interleaved_shape(*input->info())));
 
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));
 
     _input  = input;
     _output = output;
@@ -68,20 +106,9 @@ void CLGEMMInterleave4x4Kernel::configure(const ICLTensor *input, ICLTensor *out
     _kernel                 = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name));
 
     // Configure kernel window
-    const unsigned int     num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(input->info()->data_type());
-    constexpr unsigned int num_elems_processed_per_iteration_y = 4;
-    const unsigned int     num_elems_written_per_iteration     = num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y;
-
-    Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-    AccessWindowRectangle input_access(input->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
-    AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_written_per_iteration, 1, 4.f, 0.25f);
-
-    update_window_and_padding(win, input_access, output_access);
-
-    output_access.set_valid_region(win, input->info()->valid_region());
-
-    ICLKernel::configure(win);
+    auto win_config = validate_and_configure_window(input->info(), output->info());
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
 
     // Set config_id for enabling LWS tuning
     _config_id = "interleave4x4_";
@@ -92,6 +119,14 @@ void CLGEMMInterleave4x4Kernel::configure(const ICLTensor *input, ICLTensor *out
     _config_id += support::cpp11::to_string(output->info()->dimension(1));
 }
 
+Status CLGEMMInterleave4x4Kernel::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
+
+    return Status{};
+}
+
 void CLGEMMInterleave4x4Kernel::run(const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp
index 1d9fe4bc01..423592b79c 100644
--- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp
@@ -46,79 +46,114 @@ namespace arm_compute
 class Coordinates;
 } // namespace arm_compute
 
-CLGEMMLowpMatrixMultiplyKernel::CLGEMMLowpMatrixMultiplyKernel()
-    : _input0(nullptr), _input1(nullptr), _output(nullptr)
+namespace
 {
-}
+using ElementsProcessed = Steps;
 
-void CLGEMMLowpMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, bool is_interleaved_transposed)
+Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, bool is_interleaved_transposed)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::QASYMM8);
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::QASYMM8);
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
-
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::QASYMM8);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::QASYMM8);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
     if(!is_interleaved_transposed)
     {
-        ARM_COMPUTE_ERROR_ON(input0->info()->dimension(0) != input1->info()->dimension(1));
+        ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(0) != input1->dimension(1));
     }
 
-    _input0 = input0;
-    _input1 = input1;
-    _output = output;
+    return Status{};
+}
 
-    CLBuildOptions build_opts;
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output, bool is_interleaved_transposed,
+                                                        ElementsProcessed &num_elements_processed)
+{
+    unsigned int &num_elems_processed_per_iteration_x = num_elements_processed[0];
+    unsigned int &num_elems_processed_per_iteration_y = num_elements_processed[1];
+
+    Window win{};
+    bool   window_changed = false;
 
+    // Check if the output tensor is a vector. If so,the kernel runs the vector-matrix multiplication
     if(is_interleaved_transposed)
     {
-        // Create kernel and set static arguments
-        build_opts.add_option("-DCOLS_B=" + support::cpp11::to_string(input1->info()->dimension(0)));
-        _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_mm_interleaved_transposed", build_opts.options()));
-
         // Configure window
-        constexpr unsigned int num_elems_processed_per_iteration_x = 16;
-        constexpr unsigned int num_elems_processed_per_iteration_y = 4;
+        num_elems_processed_per_iteration_x                        = 16;
+        num_elems_processed_per_iteration_y                        = 4;
         constexpr unsigned int num_elems_read_per_iteration_input0 = 4;
         constexpr unsigned int num_elems_read_per_iteration_input1 = 16;
 
-        Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-        AccessWindowRectangle input0_access(input0->info(), 0, 0, num_elems_read_per_iteration_input0, 1);
-        AccessWindowRectangle input1_access(input1->info(), 0, 0, num_elems_read_per_iteration_input1, 1);
-        AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
+        win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
 
-        update_window_and_padding(win, input0_access, input1_access, output_access);
+        AccessWindowRectangle input0_access(input0, 0, 0, num_elems_read_per_iteration_input0, 1);
+        AccessWindowRectangle input1_access(input1, 0, 0, num_elems_read_per_iteration_input1, 1);
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
 
-        output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->info()->tensor_shape()));
+        window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
 
-        ICLKernel::configure(win);
+        output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->tensor_shape()));
     }
     else
     {
         // Special case for 1xN, 2xN, 3xN and 4xN input0 tensor. num_elems_processed_per_iteration_x
-        constexpr unsigned int num_elems_processed_per_iteration_x = 16;
-        const unsigned int     num_elems_processed_per_iteration_y = std::min(static_cast<int>(output->info()->dimension(1)), 4);
-
-        build_opts.add_option("-DCOLS_A=" + support::cpp11::to_string(input0->info()->dimension(0)));
-        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_X=" + support::cpp11::to_string(num_elems_processed_per_iteration_x));
-        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_Y=" + support::cpp11::to_string(num_elems_processed_per_iteration_y));
-
-        _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_mm", build_opts.options()));
+        num_elems_processed_per_iteration_x = 16;
+        num_elems_processed_per_iteration_y = std::min(static_cast<int>(output->dimension(1)), 4);
 
         // Configure window
-        Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
+        win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
 
-        AccessWindowStatic    input0_access(input0->info(), 0, 0, input0->info()->dimension(0), ceil_to_multiple(input0->info()->dimension(1), num_elems_processed_per_iteration_y));
-        AccessWindowStatic    input1_access(input1->info(), 0, 0, ceil_to_multiple(input1->info()->dimension(0), num_elems_processed_per_iteration_x), input1->info()->dimension(1));
-        AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
+        AccessWindowStatic    input0_access(input0, 0, 0, input0->dimension(0), ceil_to_multiple(input0->dimension(1), num_elems_processed_per_iteration_y));
+        AccessWindowStatic    input1_access(input1, 0, 0, ceil_to_multiple(input1->dimension(0), num_elems_processed_per_iteration_x), input1->dimension(1));
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
 
-        update_window_and_padding(win, input0_access, input1_access, output_access);
+        window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
 
         Coordinates coord;
-        coord.set_num_dimensions(output->info()->num_dimensions());
-        output_access.set_valid_region(win, ValidRegion(coord, output->info()->tensor_shape()));
+        coord.set_num_dimensions(output->num_dimensions());
+        output_access.set_valid_region(win, ValidRegion(coord, output->tensor_shape()));
+    }
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
+CLGEMMLowpMatrixMultiplyKernel::CLGEMMLowpMatrixMultiplyKernel()
+    : _input0(nullptr), _input1(nullptr), _output(nullptr)
+{
+}
+
+void CLGEMMLowpMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, bool is_interleaved_transposed)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info(), is_interleaved_transposed));
+
+    _input0 = input0;
+    _input1 = input1;
+    _output = output;
+
+    ElementsProcessed num_elements_processed{};
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info(), is_interleaved_transposed, num_elements_processed);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
 
-        ICLKernel::configure(win);
+    // Create build options
+    CLBuildOptions build_opts;
+    std::string    kernel_name(" ");
+    if(is_interleaved_transposed)
+    {
+        build_opts.add_option("-DCOLS_B=" + support::cpp11::to_string(input1->info()->dimension(0)));
+        kernel_name = "gemmlowp_mm_interleaved_transposed";
     }
+    else
+    {
+        build_opts.add_option("-DCOLS_A=" + support::cpp11::to_string(input0->info()->dimension(0)));
+        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_X=" + support::cpp11::to_string(num_elements_processed.x()));
+        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_Y=" + support::cpp11::to_string(num_elements_processed.y()));
+        kernel_name = "gemmlowp_mm";
+    }
+    // Create kernel
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options()));
 
     // Set config_id for enabling LWS tuning
     _config_id = "gemmlowp_";
@@ -132,6 +167,20 @@ void CLGEMMLowpMatrixMultiplyKernel::configure(const ICLTensor *input0, const IC
     _config_id += (is_interleaved_transposed ? support::cpp11::to_string(input1->info()->dimension(0)) : support::cpp11::to_string(input1->info()->dimension(1)));
 }
 
+Status CLGEMMLowpMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, bool is_interleaved_transposed)
+{
+    ElementsProcessed num_elements_processed{};
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output, is_interleaved_transposed));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input0->clone().get(),
+                                                              input1->clone().get(),
+                                                              output->clone().get(),
+                                                              is_interleaved_transposed,
+                                                              num_elements_processed)
+                                .first);
+
+    return Status{};
+}
+
 void CLGEMMLowpMatrixMultiplyKernel::run(const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
diff --git a/src/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.cpp b/src/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.cpp
index 2877a74be8..d05939fcf5 100644
--- a/src/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.cpp
@@ -44,89 +44,135 @@ namespace arm_compute
 class Coordinates;
 } // namespace arm_compute
 
-CLGEMMLowpOffsetContributionKernel::CLGEMMLowpOffsetContributionKernel()
-    : _vector_sum_col(nullptr), _vector_sum_row(nullptr), _mm_result(nullptr)
+namespace
 {
-}
-
-void CLGEMMLowpOffsetContributionKernel::configure(ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, int32_t k, int32_t a_offset, int32_t b_offset)
+Status validate_arguments(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row,
+                          int32_t a_offset, int32_t b_offset)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(mm_result, 1, DataType::S32);
-
-    // Set the arguments to pass at compile time
-    CLBuildOptions build_opts;
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(mm_result, 1, DataType::S32);
 
     // If a_offset == 0, vector_sum_col can be a nullptr
     if(a_offset != 0)
     {
-        ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_col, 1, DataType::S32);
-        ARM_COMPUTE_ERROR_ON(vector_sum_col->info()->dimension(0) != mm_result->info()->dimension(0));
-
-        build_opts.add_option("-DA_OFFSET=" + support::cpp11::to_string(a_offset));
-        build_opts.add_option_if(vector_sum_col->info()->tensor_shape().num_dimensions() > 1, "-DSUM_COL_HAS_BATCHES");
+        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_col, 1, DataType::S32);
+        ARM_COMPUTE_RETURN_ERROR_ON(vector_sum_col->dimension(0) != mm_result->dimension(0));
     }
 
     // If b_offset == 0, vector_sum_row can be a nullptr
     if(b_offset != 0)
     {
-        ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_row, 1, DataType::S32);
-        ARM_COMPUTE_ERROR_ON(vector_sum_row->info()->dimension(0) != mm_result->info()->dimension(1));
+        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_row, 1, DataType::S32);
+        ARM_COMPUTE_RETURN_ERROR_ON(vector_sum_row->dimension(0) != mm_result->dimension(1));
 
-        // Validate batches
-        TensorShape output_shape = mm_result->info()->tensor_shape();
+        TensorShape output_shape = mm_result->tensor_shape();
         if(output_shape.num_dimensions() > 1)
         {
-            TensorShape vector_sum_row_shape = vector_sum_row->info()->tensor_shape();
+            TensorShape vector_sum_row_shape = vector_sum_row->tensor_shape();
             vector_sum_row_shape.collapse_from(1);
             output_shape.collapse_from(2);
 
-            ARM_COMPUTE_ERROR_ON_MSG(vector_sum_row_shape[1] != output_shape[2], "mm_result tensor must have the same number of batches of output tensor");
+            ARM_COMPUTE_RETURN_ERROR_ON_MSG(vector_sum_row_shape[1] != output_shape[2],
+                                            "mm_result tensor must have the same number of batches of output tensor");
 
             if(a_offset != 0)
             {
-                TensorShape vector_sum_col_shape = vector_sum_col->info()->tensor_shape();
+                TensorShape vector_sum_col_shape = vector_sum_col->tensor_shape();
                 vector_sum_col_shape.collapse_from(1);
 
-                ARM_COMPUTE_ERROR_ON_MSG(vector_sum_col_shape[1] != 1
-                                         && vector_sum_col_shape[1] != vector_sum_row_shape[1],
-                                         "vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1");
+                ARM_COMPUTE_RETURN_ERROR_ON_MSG(vector_sum_col_shape[1] != 1 && vector_sum_col_shape[1] != vector_sum_row_shape[1],
+                                                "vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1");
             }
         }
-
-        build_opts.add_option("-DB_OFFSET=" + support::cpp11::to_string(b_offset));
     }
 
-    build_opts.add_option("-DK_OFFSET=" + support::cpp11::to_string(a_offset * b_offset * k));
-
-    // Create kernel
-    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_offset_contribution", build_opts.options()));
-
-    _vector_sum_col = vector_sum_col;
-    _vector_sum_row = vector_sum_row;
-    _mm_result      = mm_result;
+    return Status{};
+}
 
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *mm_result, ITensorInfo *vector_sum_col, ITensorInfo *vector_sum_row,
+                                                        int32_t a_offset, int32_t b_offset)
+{
     constexpr unsigned int num_elems_processed_per_iteration = 16;
+    bool                   window_changed                    = false;
 
     // Configure kernel window
-    Window win = calculate_max_window(*mm_result->info(), Steps(num_elems_processed_per_iteration));
+    Window win = calculate_max_window(*mm_result, Steps(num_elems_processed_per_iteration));
 
-    AccessWindowHorizontal mm_result_access(mm_result->info(), 0, num_elems_processed_per_iteration);
-
-    update_window_and_padding(win, mm_result_access);
+    AccessWindowHorizontal mm_result_access(mm_result, 0, num_elems_processed_per_iteration);
+    window_changed = window_changed || update_window_and_padding(win,
+                                                                 mm_result_access);
 
     if(a_offset != 0)
     {
-        AccessWindowHorizontal vector_sum_col_access(vector_sum_col->info(), 0, num_elems_processed_per_iteration);
-        update_window_and_padding(win, vector_sum_col_access);
+        AccessWindowHorizontal vector_sum_col_access(vector_sum_col, 0, num_elems_processed_per_iteration);
+        window_changed = window_changed || update_window_and_padding(win,
+                                                                     vector_sum_col_access);
     }
-
     if(b_offset != 0)
     {
-        AccessWindowStatic vector_sum_row_access(vector_sum_row->info(), 0, 0, vector_sum_row->info()->dimension(0), 0);
-        update_window_and_padding(win, vector_sum_row_access);
+        AccessWindowStatic vector_sum_row_access(vector_sum_row, 0, 0, vector_sum_row->dimension(0), 0); // NOLINT
+        window_changed = window_changed || update_window_and_padding(win,
+                                                                     vector_sum_row_access);
     }
 
-    ICLKernel::configure(win);
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
+CLGEMMLowpOffsetContributionKernel::CLGEMMLowpOffsetContributionKernel()
+    : _vector_sum_col(nullptr), _vector_sum_row(nullptr), _mm_result(nullptr)
+{
+}
+
+void CLGEMMLowpOffsetContributionKernel::configure(ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, int32_t k, int32_t a_offset, int32_t b_offset)
+{
+    // Perform validate step
+    ARM_COMPUTE_ERROR_ON_NULLPTR(mm_result);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(mm_result->info(),
+                                                  vector_sum_col != nullptr ? vector_sum_col->info() : nullptr,
+                                                  vector_sum_row != nullptr ? vector_sum_row->info() : nullptr,
+                                                  a_offset, b_offset)); // NOLINT
+
+    _vector_sum_col = vector_sum_col;
+    _vector_sum_row = vector_sum_row;
+    _mm_result      = mm_result;
+
+    // Set the arguments to pass at compile time
+    CLBuildOptions build_opts;
+
+    // If a_offset == 0, vector_sum_col can be a nullptr
+    if(a_offset != 0)
+    {
+        build_opts.add_option("-DA_OFFSET=" + support::cpp11::to_string(a_offset));
+        build_opts.add_option_if(vector_sum_col->info()->tensor_shape().num_dimensions() > 1, "-DSUM_COL_HAS_BATCHES");
+    }
+    // If b_offset == 0, vector_sum_row can be a nullptr
+    build_opts.add_option_if(b_offset != 0, "-DB_OFFSET=" + support::cpp11::to_string(b_offset));
+    build_opts.add_option("-DK_OFFSET=" + support::cpp11::to_string(a_offset * b_offset * k));
+
+    // Create kernel
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_offset_contribution", build_opts.options()));
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(mm_result->info(),
+                                                    vector_sum_col != nullptr ? vector_sum_col->info() : nullptr,
+                                                    vector_sum_row != nullptr ? vector_sum_row->info() : nullptr,
+                                                    a_offset, b_offset); // NOLINT
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
+}
+
+Status CLGEMMLowpOffsetContributionKernel::validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row,
+                                                    int32_t a_offset, int32_t b_offset)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(mm_result, vector_sum_col, vector_sum_row, a_offset, b_offset));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(mm_result->clone().get(),
+                                                              vector_sum_col != nullptr ? vector_sum_col->clone().get() : nullptr,
+                                                              vector_sum_row != nullptr ? vector_sum_row->clone().get() : nullptr,
+                                                              a_offset, b_offset)
+                                .first); // NOLINT
+
+    return Status{};
 }
 
 void CLGEMMLowpOffsetContributionKernel::run(const Window &window, cl::CommandQueue &queue)
diff --git a/src/core/CL/kernels/CLGEMMLowpReductionKernel.cpp b/src/core/CL/kernels/CLGEMMLowpReductionKernel.cpp
index bcf04b0982..6951512167 100644
--- a/src/core/CL/kernels/CLGEMMLowpReductionKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMLowpReductionKernel.cpp
@@ -44,6 +44,59 @@ namespace arm_compute
 class Coordinates;
 } // namespace arm_compute
 
+namespace
+{
+Status validate_arguments_matrix_a_reduction(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
+
+    return Status{};
+}
+std::pair<Status, Window> validate_and_configure_window_matrix_a_reduction(ITensorInfo *input, ITensorInfo *output)
+{
+    const unsigned int num_elems_processed_per_iteration = 1;
+
+    Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));
+
+    AccessWindowStatic     input_access(input, 0, 0, ceil_to_multiple(input->dimension(0), 16), input->dimension(1));
+    AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
+
+    bool window_changed = update_window_and_padding(win, input_access, output_access);
+
+    output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->tensor_shape()));
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+
+Status validate_arguments_matrix_b_reduction(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window_matrix_b_reduction(ITensorInfo *input, ITensorInfo *output)
+{
+    constexpr unsigned int num_elems_processed_per_iteration = 16;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration));
+
+    AccessWindowStatic     input_access(input, 0, 0, ceil_to_multiple(input->dimension(0), num_elems_processed_per_iteration), input->dimension(1));
+    AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
+
+    bool window_changed = update_window_and_padding(win, input_access, output_access);
+
+    output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->tensor_shape()));
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
 ICLGEMMLowpReductionKernel::ICLGEMMLowpReductionKernel()
     : _input(), _output()
 {
@@ -51,8 +104,9 @@ ICLGEMMLowpReductionKernel::ICLGEMMLowpReductionKernel()
 
 void CLGEMMLowpMatrixAReductionKernel::configure(const ICLTensor *mtx_a, ICLTensor *vector_sum_row)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(mtx_a, 1, DataType::QASYMM8);
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_row, 1, DataType::S32);
+    // Perform validate step
+    ARM_COMPUTE_ERROR_ON_NULLPTR(mtx_a, vector_sum_row);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_matrix_a_reduction(mtx_a->info(), vector_sum_row->info()));
 
     _input  = mtx_a;
     _output = vector_sum_row;
@@ -64,21 +118,18 @@ void CLGEMMLowpMatrixAReductionKernel::configure(const ICLTensor *mtx_a, ICLTens
     // Create kernel
     _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_matrix_a_reduction", build_opts.options()));
 
-    const unsigned int num_elems_processed_per_iteration = 1;
-
     // Configure kernel window
-    Window win = calculate_max_window(*_output->info(), Steps(num_elems_processed_per_iteration));
-
-    AccessWindowStatic     input_access(_input->info(), 0, 0, ceil_to_multiple(_input->info()->dimension(0), 16), _input->info()->dimension(1));
-    AccessWindowHorizontal output_access(_output->info(), 0, num_elems_processed_per_iteration);
-
-    update_window_and_padding(win,
-                              input_access,
-                              output_access);
+    auto win_config = validate_and_configure_window_matrix_a_reduction(_input->info(), _output->info());
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
+}
 
-    output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), _output->info()->tensor_shape()));
+Status CLGEMMLowpMatrixAReductionKernel::validate(const ITensorInfo *mtx_a, const ITensorInfo *vector_sum_row)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_matrix_a_reduction(mtx_a, vector_sum_row));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_matrix_a_reduction(mtx_a->clone().get(), vector_sum_row->clone().get()).first);
 
-    ICLKernel::configure(win);
+    return Status{};
 }
 
 void CLGEMMLowpMatrixAReductionKernel::run(const Window &window, cl::CommandQueue &queue)
@@ -107,8 +158,8 @@ void CLGEMMLowpMatrixAReductionKernel::run(const Window &window, cl::CommandQueu
 
 void CLGEMMLowpMatrixBReductionKernel::configure(const ICLTensor *mtx_b, ICLTensor *vector_sum_col)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(mtx_b, 1, DataType::QASYMM8);
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(vector_sum_col, 1, DataType::S32);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(mtx_b, vector_sum_col);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_matrix_b_reduction(mtx_b->info(), vector_sum_col->info()));
 
     _input  = mtx_b;
     _output = vector_sum_col;
@@ -121,21 +172,18 @@ void CLGEMMLowpMatrixBReductionKernel::configure(const ICLTensor *mtx_b, ICLTens
     // Create kernel
     _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_matrix_b_reduction", build_opts.options()));
 
-    constexpr unsigned int num_elems_processed_per_iteration = 16;
-
     // Configure kernel window
-    Window win = calculate_max_window(*vector_sum_col->info(), Steps(num_elems_processed_per_iteration));
-
-    AccessWindowStatic     input_access(_input->info(), 0, 0, ceil_to_multiple(_input->info()->dimension(0), num_elems_processed_per_iteration), _input->info()->dimension(1));
-    AccessWindowHorizontal output_access(_output->info(), 0, num_elems_processed_per_iteration);
-
-    update_window_and_padding(win,
-                              input_access,
-                              output_access);
+    auto win_config = validate_and_configure_window_matrix_b_reduction(_input->info(), _output->info());
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
+}
 
-    output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), _output->info()->tensor_shape()));
+Status CLGEMMLowpMatrixBReductionKernel::validate(const ITensorInfo *mtx_b, const ITensorInfo *vector_sum_col)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_matrix_b_reduction(mtx_b, vector_sum_col));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_matrix_b_reduction(mtx_b->clone().get(), vector_sum_col->clone().get()).first);
 
-    ICLKernel::configure(win);
+    return Status{};
 }
 
 void CLGEMMLowpMatrixBReductionKernel::run(const Window &window, cl::CommandQueue &queue)
diff --git a/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp
index 015b4f70a4..d5c93dd24a 100644
--- a/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.cpp
@@ -36,6 +36,37 @@
 
 using namespace arm_compute;
 
+namespace
+{
+Status validate_arguments(const ITensorInfo *accum, const ITensorInfo *biases)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(accum, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(biases, accum);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(biases, accum);
+    ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() != 1);
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *accum, ITensorInfo *biases, GPUTarget gpu_target,
+                                                        unsigned int &num_elems_processed_per_iteration)
+{
+    // Select the vector size to use (8 for Bifrost; 16 for Midgard).
+    num_elems_processed_per_iteration = (gpu_target == GPUTarget::BIFROST) ? 8 : 16;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*accum, Steps(num_elems_processed_per_iteration));
+
+    AccessWindowStatic     biases_access(biases, 0, 0, ceil_to_multiple(biases->dimension(0), num_elems_processed_per_iteration), biases->dimension(1));
+    AccessWindowHorizontal accum_access(accum, 0, num_elems_processed_per_iteration);
+
+    bool window_changed = update_window_and_padding(win, biases_access, accum_access);
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
 CLGEMMMatrixAccumulateBiasesKernel::CLGEMMMatrixAccumulateBiasesKernel()
     : _accum(nullptr), _biases(nullptr)
 {
@@ -43,18 +74,21 @@ CLGEMMMatrixAccumulateBiasesKernel::CLGEMMMatrixAccumulateBiasesKernel()
 
 void CLGEMMMatrixAccumulateBiasesKernel::configure(ICLTensor *accum, const ICLTensor *biases)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(accum, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(biases, accum);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(biases, accum);
-    ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() != 1);
+    // Perform validate step
+    ARM_COMPUTE_ERROR_ON_NULLPTR(accum, biases);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(accum->info(), biases->info()));
 
     _biases = biases;
     _accum  = accum;
 
     // Get the target architecture
-    GPUTarget arch_target = get_arch_from_target(get_target());
-    // Select the vector size to use (8 for Bifrost; 16 for Midgard).
-    const unsigned int vector_size = (arch_target == GPUTarget::BIFROST) ? 8 : 16;
+    GPUTarget    arch_target = get_arch_from_target(get_target());
+    unsigned int vector_size = 0;
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(accum->info(), biases->info(), arch_target, vector_size);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
 
     // Add build options
     CLBuildOptions build_opts;
@@ -65,18 +99,15 @@ void CLGEMMMatrixAccumulateBiasesKernel::configure(ICLTensor *accum, const ICLTe
 
     // Create kernel
     _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemm_accumulate_biases", build_opts.options()));
+}
 
-    // Configure kernel window
-    const unsigned int num_elems_processed_per_iteration = vector_size;
-
-    Window win = calculate_max_window(*_accum->info(), Steps(num_elems_processed_per_iteration));
-
-    AccessWindowStatic     biases_access(biases->info(), 0, 0, ceil_to_multiple(biases->info()->dimension(0), num_elems_processed_per_iteration), biases->info()->dimension(1));
-    AccessWindowHorizontal accum_access(_accum->info(), 0, num_elems_processed_per_iteration);
-
-    update_window_and_padding(win, biases_access, accum_access);
+Status CLGEMMMatrixAccumulateBiasesKernel::validate(const ITensorInfo *accum, const ITensorInfo *biases, GPUTarget gpu_target)
+{
+    unsigned int num_elems_processed_per_iteration = 0;
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(accum, biases));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(accum->clone().get(), biases->clone().get(), gpu_target, num_elems_processed_per_iteration).first);
 
-    ICLKernel::configure(win);
+    return Status{};
 }
 
 void CLGEMMMatrixAccumulateBiasesKernel::run(const Window &window, cl::CommandQueue &queue)
diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp
index 16706dd748..f51d0f92d4 100644
--- a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp
@@ -42,6 +42,81 @@
 
 using namespace arm_compute;
 
+namespace
+{
+using ElementsProcessed = Steps;
+
+inline Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, bool is_interleaved_transposed)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input0, input1, output);
+    if(!is_interleaved_transposed)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(0) != input1->dimension(1));
+    }
+
+    return Status{};
+}
+
+inline std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output,
+                                                               bool is_interleaved_transposed, GPUTarget gpu_target,
+                                                               ElementsProcessed &num_elements_processed)
+{
+    bool   window_changed = false;
+    Window win{};
+
+    const DataType data_type                           = input0->data_type();
+    unsigned int &num_elems_processed_per_iteration_x = num_elements_processed[0];
+    unsigned int &num_elems_processed_per_iteration_y = num_elements_processed[1];
+
+    if(is_interleaved_transposed)
+    {
+        // Configure kernel window
+        num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(data_type);
+        num_elems_processed_per_iteration_y = 4;
+
+        win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
+
+        AccessWindowRectangle input0_access(input0, 0, 0, num_elems_processed_per_iteration_y, 1, 1.f, 0.25f);
+        AccessWindowTranspose input1_access(input1, 0, 0, num_elems_processed_per_iteration_x, 1, 0.f, 0.25f);
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
+
+        window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
+
+        output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->tensor_shape()));
+    }
+    else // The input tensors have not been reshaped
+    {
+        // Special case for 1xN, 2xN, 3xN and 4xN input0 tensor. num_elems_processed_per_iteration_x is set up for the default case.
+        num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(data_type);
+        num_elems_processed_per_iteration_y = std::min(static_cast<int>(output->dimension(1)), 4);
+
+        // Create kernels according to the architecture, data type and input size.
+        if(gpu_target == GPUTarget::BIFROST && data_type == DataType::F32)
+        {
+            num_elems_processed_per_iteration_x = (input1->dimension(0) <= 1000) ? 2 : 4;
+        }
+
+        // Configure window
+        win = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
+
+        AccessWindowStatic    input0_access(input0, 0, 0, input0->dimension(0), ceil_to_multiple(input0->dimension(1), num_elems_processed_per_iteration_y));
+        AccessWindowStatic    input1_access(input1, 0, 0, ceil_to_multiple(input1->dimension(0), num_elems_processed_per_iteration_x), input1->dimension(1));
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
+
+        window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
+
+        Coordinates coord;
+        coord.set_num_dimensions(output->num_dimensions());
+        output_access.set_valid_region(win, ValidRegion(coord, output->tensor_shape()));
+    }
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
 CLGEMMMatrixMultiplyKernel::CLGEMMMatrixMultiplyKernel()
     : _input0(nullptr), _input1(nullptr), _output(nullptr)
 {
@@ -49,13 +124,10 @@ CLGEMMMatrixMultiplyKernel::CLGEMMMatrixMultiplyKernel()
 
 void CLGEMMMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, float alpha, bool is_interleaved_transposed)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input0, input1, output);
-    if(!is_interleaved_transposed)
-    {
-        ARM_COMPUTE_ERROR_ON(input0->info()->dimension(0) != input1->info()->dimension(1));
-    }
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
+
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info(), is_interleaved_transposed));
 
     _input0 = input0;
     _input1 = input1;
@@ -82,14 +154,19 @@ void CLGEMMMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTen
         _lws_hint = cl::NDRange(8, 8);
     }
 
+    ElementsProcessed num_elements_processed{};
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info(), is_interleaved_transposed, arch_target, num_elements_processed);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
+
     // Create build options
     CLBuildOptions build_opts;
     build_opts.add_option_if(is_data_type_fixed_point(data_type), "-DFIXED_POINT_POSITION=" + support::cpp11::to_string(fp_pos));
 
-    const bool multiply_alpha = std::abs(1.0f - alpha) > 0.00001f;
-
     // Only define ALPHA when alpha is not 1.0f. This avoids performing unnecessary multiplications.
-    if(multiply_alpha)
+    if(std::abs(1.0f - alpha) > 0.00001f)
     {
         build_opts.add_option_if_else(is_data_type_fixed_point(data_type),
                                       "-DALPHA=" + support::cpp11::to_string((data_type == DataType::QS8 ? sqcvt_qs8_f32(alpha, fp_pos) : sqcvt_qs16_f32(alpha, fp_pos))),
@@ -108,49 +185,19 @@ void CLGEMMMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTen
         {
             kernel_name = "gemm_mm_interleaved_transposed_" + lower_string(string_from_data_type(data_type));
         }
-
-        // Configure kernel window
-        const unsigned int     num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(data_type);
-        constexpr unsigned int num_elems_processed_per_iteration_y = 4;
-
-        Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-        AccessWindowRectangle input0_access(input0->info(), 0, 0, num_elems_processed_per_iteration_y, 1, 1.f, 0.25f);
-        AccessWindowTranspose input1_access(input1->info(), 0, 0, num_elems_processed_per_iteration_x, 1, 0.f, 0.25f);
-        AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
-
-        update_window_and_padding(win, input0_access, input1_access, output_access);
-
-        output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->info()->tensor_shape()));
-
-        ICLKernel::configure(win);
     }
     else // The input tensors have not been reshaped
     {
         build_opts.add_option("-DCOLS_A=" + support::cpp11::to_string(input0->info()->dimension(0)));
 
-        // Special case for 1xN, 2xN, 3xN and 4xN input0 tensor. num_elems_processed_per_iteration_x is set up for the default case.
-        unsigned int       num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(data_type);
-        const unsigned int num_elems_processed_per_iteration_y = std::min(static_cast<int>(output->info()->dimension(1)), 4);
-
         // Create kernels according to the architecture, data type and input size.
         if(arch_target == GPUTarget::BIFROST && data_type == DataType::F32)
         {
             // The first kernel is optimized for the case of 1000 or less output elements (e.g. FC8 of AlexNet and VGG-16, and
             // FC1 of Inception v3). The second kernel is optimized for the case of greater than 1000 output elements (e.g.
             // FC6 and FC7 of AlexNet and VGG-16).
-            if(input1->info()->dimension(0) <= 1000)
-            {
-                // Each work-item processes 2 elements in the X dimension.
-                num_elems_processed_per_iteration_x = 2;
-                kernel_name                         = "gemm_mm_floating_point_f32_bifrost_1000";
-            }
-            else
-            {
-                // Each work-item processes 4 elements in the X dimension (as in the default case).
-                num_elems_processed_per_iteration_x = 4;
-                kernel_name                         = "gemm_mm_floating_point_f32_bifrost";
-            }
+            kernel_name = (input1->info()->dimension(0) <= 1000) ? "gemm_mm_floating_point_f32_bifrost_1000" : "gemm_mm_floating_point_f32_bifrost";
+
             // The work-group size equal to the Bifrost quad size has been proved to be optimal for these kernels
             // via exhaustive autotuning over a range of representative layer configurations.
             _lws_hint = cl::NDRange(4);
@@ -164,23 +211,8 @@ void CLGEMMMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTen
             build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
             kernel_name = "gemm_mm_floating_point";
         }
-        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_Y=" + support::cpp11::to_string(num_elems_processed_per_iteration_y));
-        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_X=" + support::cpp11::to_string(num_elems_processed_per_iteration_x));
-
-        // Configure window
-        Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-        AccessWindowStatic    input0_access(input0->info(), 0, 0, input0->info()->dimension(0), ceil_to_multiple(input0->info()->dimension(1), num_elems_processed_per_iteration_y));
-        AccessWindowStatic    input1_access(input1->info(), 0, 0, ceil_to_multiple(input1->info()->dimension(0), num_elems_processed_per_iteration_x), input1->info()->dimension(1));
-        AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
-
-        update_window_and_padding(win, input0_access, input1_access, output_access);
-
-        Coordinates coord;
-        coord.set_num_dimensions(output->info()->num_dimensions());
-        output_access.set_valid_region(win, ValidRegion(coord, output->info()->tensor_shape()));
-
-        ICLKernel::configure(win);
+        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_Y=" + support::cpp11::to_string(num_elements_processed.y()));
+        build_opts.add_option("-DNUM_ELEMS_PROCESSED_PER_THREAD_X=" + support::cpp11::to_string(num_elements_processed.x()));
     }
 
     // Create kernel
@@ -198,6 +230,22 @@ void CLGEMMMatrixMultiplyKernel::configure(const ICLTensor *input0, const ICLTen
     _config_id += (is_interleaved_transposed ? support::cpp11::to_string(input1->info()->dimension(0)) : support::cpp11::to_string(input1->info()->dimension(1)));
 }
 
+Status CLGEMMMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved_transposed, GPUTarget gpu_target)
+{
+    ElementsProcessed num_elements_processed{};
+    ARM_COMPUTE_UNUSED(alpha);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output, is_interleaved_transposed));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input0->clone().get(),
+                                                              input1->clone().get(),
+                                                              output->clone().get(),
+                                                              is_interleaved_transposed,
+                                                              gpu_target,
+                                                              num_elements_processed)
+                                .first);
+
+    return Status{};
+}
+
 void CLGEMMMatrixMultiplyKernel::run(const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
diff --git a/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp b/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp
index 35074f94cf..69a545b76b 100644
--- a/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp
@@ -33,36 +33,82 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 #include <cmath>
 
 using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
 
-void CLGEMMTranspose1xWKernel::configure(const ICLTensor *input, ICLTensor *output)
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::QASYMM8,
-                                                  DataType::U16, DataType::S16, DataType::QS16,
-                                                  DataType::U32, DataType::S32, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_NULLPTR(output);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::U8, DataType::S8,
+                                                         DataType::QS16, DataType::U16, DataType::S16, DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(),
+                                                           compute_transpose1xW_with_element_size_shape(*input));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    }
 
-    TensorShape  output_shape{ input->info()->tensor_shape() };
-    const size_t transpose_w = 16 / input->info()->element_size();
-    output_shape.set(0, input->info()->dimension(1) * transpose_w);
-    output_shape.set(1, static_cast<size_t>(std::ceil((input->info()->dimension(0) / static_cast<float>(transpose_w)))));
+    return Status{};
+}
 
-    // Output tensor auto inizialitation if not yet initialized
-    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, unsigned int &num_elems_processed_per_iteration)
+{
+    num_elems_processed_per_iteration = 16 / input->element_size();
+
+    const int scale_x        = num_elems_processed_per_iteration;
+    bool      window_changed = false;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration));
+
+    if((win.x().end() / scale_x) == 0)
+    {
+        return std::make_pair(ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Transposed shape would be 0 in the second dimension"), win);
+    }
+
+    AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration);
+    window_changed = window_changed || update_window_and_padding(win, input_access);
+
+    // Configure window in case of configured output
+    if(output->total_size() != 0)
+    {
+        AccessWindowTranspose output_access(output, 0, 0, num_elems_processed_per_iteration, 1, scale_x, 1.f / scale_x);
+        window_changed = window_changed || update_window_and_padding(win, output_access);
+        output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), input->tensor_shape()));
+    }
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
 
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+void CLGEMMTranspose1xWKernel::configure(const ICLTensor *input, ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    // Output tensor auto inizialitation if not yet initialized
+    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(compute_transpose1xW_with_element_size_shape(*input->info())));
 
-    const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size();
-    const int          scale_x                           = num_elems_processed_per_iteration;
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));
 
     _input  = input;
     _output = output;
 
+    // Configure kernel window
+    unsigned int num_elems_processed_per_iteration = 1;
+    auto         win_config                        = validate_and_configure_window(input->info(), output->info(), num_elems_processed_per_iteration);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure(win_config.second);
+
     /*
      * Following an example of how the transposition1xW works when the input data type is F32
      *
@@ -76,20 +122,15 @@ void CLGEMMTranspose1xWKernel::configure(const ICLTensor *input, ICLTensor *outp
     // Create kernel
     std::string kernel_name = "gemm_transpose1x" + support::cpp11::to_string(num_elems_processed_per_iteration);
     _kernel                 = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name));
+}
 
-    // Configure window
-    Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration));
-
-    ARM_COMPUTE_ERROR_ON_MSG((win.x().end() / scale_x) == 0, "Transposed shape would be 0 in the second dimension");
-
-    AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration);
-    AccessWindowTranspose  output_access(output->info(), 0, 0, num_elems_processed_per_iteration, 1, scale_x, 1.f / scale_x);
-
-    update_window_and_padding(win, input_access, output_access);
-
-    output_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), input->info()->tensor_shape()));
+Status CLGEMMTranspose1xWKernel::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+    unsigned int num_elems_processed_per_iteration = 1;
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), num_elems_processed_per_iteration).first);
 
-    ICLKernel::configure(win);
+    return Status{};
 }
 
 void CLGEMMTranspose1xWKernel::run(const Window &window, cl::CommandQueue &queue)
diff --git a/src/core/CL/kernels/CLIm2ColKernel.cpp b/src/core/CL/kernels/CLIm2ColKernel.cpp
index 6514d6cf91..0e9f2c5344 100644
--- a/src/core/CL/kernels/CLIm2ColKernel.cpp
+++ b/src/core/CL/kernels/CLIm2ColKernel.cpp
@@ -39,6 +39,24 @@
 
 using namespace arm_compute;
 
+namespace
+{
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
+
+    // Checks performed when output is configured
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    }
+
+    return Status{};
+}
+} // namespace
+
 CLIm2ColKernel::CLIm2ColKernel()
     : _input(nullptr), _output(nullptr), _convolved_dims(), _num_elems_processed_per_iteration(1), _run_func(nullptr)
 {
@@ -46,9 +64,10 @@ CLIm2ColKernel::CLIm2ColKernel()
 
 void CLIm2ColKernel::configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::QS16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    // Perform validation step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));
 
     _input  = input;
     _output = output;
@@ -184,6 +203,15 @@ void CLIm2ColKernel::configure(const ICLTensor *input, ICLTensor *output, const
     _config_id += support::cpp11::to_string(output->info()->dimension(1));
 }
 
+Status CLIm2ColKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+{
+    ARM_COMPUTE_UNUSED(kernel_dims);
+    ARM_COMPUTE_UNUSED(conv_info);
+    ARM_COMPUTE_UNUSED(has_bias);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
+    return Status{};
+}
+
 void CLIm2ColKernel::run(const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON(_run_func == nullptr);
diff --git a/src/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.cpp
index f6964002dd..3d41548a6a 100644
--- a/src/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.cpp
@@ -71,16 +71,16 @@ Status validate_arguments(const ITensorInfo *mm_result, const ITensorInfo *vecto
             vector_sum_row_shape.collapse_from(1);
             output_shape.collapse_from(2);
 
-            ARM_COMPUTE_ERROR_ON_MSG(vector_sum_row_shape[1] != output_shape[2], "mm_result tensor must have the same number of batches of output tensor");
+            ARM_COMPUTE_RETURN_ERROR_ON_MSG(vector_sum_row_shape[1] != output_shape[2],
+                                            "mm_result tensor must have the same number of batches of output tensor");
 
             if(a_offset != 0)
             {
                 TensorShape vector_sum_col_shape = vector_sum_col->tensor_shape();
                 vector_sum_col_shape.collapse_from(1);
 
-                ARM_COMPUTE_ERROR_ON_MSG(vector_sum_col_shape[1] != 1
-                                         && vector_sum_col_shape[1] != vector_sum_row_shape[1],
-                                         "vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1");
+                ARM_COMPUTE_RETURN_ERROR_ON_MSG(vector_sum_col_shape[1] != 1 && vector_sum_col_shape[1] != vector_sum_row_shape[1],
+                                                "vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1");
             }
         }
     }
diff --git a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
index 7fd81cdb94..68c6576a79 100644
--- a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
@@ -25,6 +25,7 @@
 
 #include "arm_compute/core/Size2D.h"
 #include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/core/utils/quantization/AsymmHelpers.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "support/ToolchainSupport.h"
@@ -32,6 +33,34 @@
 #include <algorithm>
 
 using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
+
+namespace
+{
+Status validate_mm(const ITensorInfo &input, const ITensorInfo &weights, const ITensorInfo &output, bool is_interleaved_transposed)
+{
+    const GPUTarget gpu_target = CLScheduler::get().target();
+
+    if(is_data_type_quantized_asymmetric(input.data_type()))
+    {
+        // Since we need negative offsets for computing convolution, we need to change QuantizationInfo()
+        // Extract and negate input and weights offset
+        const QuantizationInfo input_quantization_info(input.quantization_info().scale, -input.quantization_info().offset);
+        const QuantizationInfo weights_quantization_info(weights.quantization_info().scale, -weights.quantization_info().offset);
+
+        // Validate gemmlowp function
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyCore::validate(&input.clone()->set_quantization_info(input_quantization_info),
+                                                                           &weights.clone()->set_quantization_info(weights_quantization_info),
+                                                                           &output));
+    }
+    else
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMMatrixMultiplyKernel::validate(&input, &weights, &output, 1.f, is_interleaved_transposed, gpu_target));
+    }
+
+    return Status{};
+}
+} // namespace
 
 void CLFullyConnectedLayerReshapeWeights::configure(const ICLTensor *input, ICLTensor *output)
 {
@@ -40,6 +69,11 @@ void CLFullyConnectedLayerReshapeWeights::configure(const ICLTensor *input, ICLT
     _kernel = std::move(k);
 }
 
+Status CLFullyConnectedLayerReshapeWeights::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+    return CLTransposeKernel::validate(input, output);
+}
+
 CLFullyConnectedLayer::CLFullyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager)
     : _memory_group(memory_manager), _im2col_kernel(), _reshape_weights_kernel(), _mm_kernel(), _mm_gemmlowp(memory_manager), _gemmlowp_output_stage(), _accumulate_biases_kernel(), _im2col_output(),
       _gemmlowp_output(), _reshape_weights_output(), _are_weights_reshaped(true), _is_fc_after_conv(true), _accumulate_biases(false), _is_quantized(false)
@@ -80,8 +114,7 @@ void CLFullyConnectedLayer::configure_conv_fc(const ICLTensor *input, const ICLT
     // If the fully connected layer is called after a convolution layer, the input tensor must be linearized
 
     // Initialize output tensor for im2col
-    TensorShape shape_im2col = input->info()->tensor_shape();
-    shape_im2col.collapse(3);
+    TensorShape shape_im2col = compute_im2col_shape(*input->info());
     _im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col));
 
     // Configure im2col kernel
@@ -105,9 +138,15 @@ void CLFullyConnectedLayer::configure_fc_fc(const ICLTensor *input, const ICLTen
 
 void CLFullyConnectedLayer::configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, bool transpose_weights, bool are_weights_reshaped)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::QS16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, output);
-    ARM_COMPUTE_ERROR_ON(weights->info()->num_dimensions() > 2);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
+
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(CLFullyConnectedLayer::validate(input->info(),
+                                                               weights->info(),
+                                                               biases != nullptr ? biases->info() : nullptr,
+                                                               output->info(),
+                                                               transpose_weights,
+                                                               are_weights_reshaped));
 
     _are_weights_reshaped = transpose_weights ? are_weights_reshaped : true;
     _is_fc_after_conv     = true;
@@ -192,6 +231,86 @@ void CLFullyConnectedLayer::configure(const ICLTensor *input, const ICLTensor *w
     }
 }
 
+Status CLFullyConnectedLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, bool transpose_weights, bool are_weights_reshaped)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, output);
+    ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 2);
+
+    bool            weights_reshaped = transpose_weights ? are_weights_reshaped : true;
+    bool            is_fc_after_conv = true;
+    bool            is_quantized     = is_data_type_quantized_asymmetric(input->data_type());
+    const GPUTarget gpu_target       = CLScheduler::get().target();
+
+    const ITensorInfo &im2col_input     = TensorInfo(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_im2col_shape(*input)));
+    const ITensorInfo &reshaped_weights = TensorInfo(weights->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_transposed_shape(*weights)));
+    const ITensorInfo &gemmlowp_output  = TensorInfo(output->clone()->set_is_resizable(true).reset_padding().set_data_type(DataType::S32));
+
+    // Configure accumulate biases kernel for non quantized asymmetric types
+    if(biases != nullptr && !is_quantized)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases);
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMMatrixAccumulateBiasesKernel::validate(output, biases, gpu_target));
+    }
+
+    // With the Fully Connected layer we can have 4 different cases:
+    //  1) Convolution layer -> Fully Connected layer without batches
+    //  2) Fully Connected layer -> Fully Connected layer without batches
+    //  3) Convolution layer -> Fully Connected layer with batches
+    //  4) Fully Connected layer -> Fully Connected layer with batches
+
+    const ITensorInfo *input_to_use   = input;
+    const ITensorInfo *weights_to_use = weights;
+    const ITensorInfo *tmp_output     = (is_quantized) ? &gemmlowp_output : output;
+
+    if(!weights_reshaped)
+    {
+        // Validate reshape weights kernel
+        ARM_COMPUTE_RETURN_ON_ERROR(CLFullyConnectedLayerReshapeWeights::validate(weights, &reshaped_weights));
+        weights_to_use = &reshaped_weights;
+    }
+
+    // Check if we have a fully connected layer with batches
+    const bool is_batched_fc_layer = output->dimension(1) > 1;
+
+    if(is_batched_fc_layer)
+    {
+        is_fc_after_conv = (TensorShape::num_max_dimensions >= 4) && (std::equal(input->tensor_shape().cbegin() + 3,
+                                                                                 input->tensor_shape().cend(),
+                                                                                 output->tensor_shape().cbegin() + 1));
+    }
+    else
+    {
+        is_fc_after_conv = input->num_dimensions() > 1;
+    }
+
+    if(is_fc_after_conv)
+    {
+        // Fully Connected layer after a Convolution Layer without batches
+        ARM_COMPUTE_RETURN_ERROR_ON((weights_to_use->dimension(1) != (input->dimension(0) * input->dimension(1) * input->dimension(2))));
+
+        // Validate im2col kernel
+        ARM_COMPUTE_RETURN_ON_ERROR(CLIm2ColKernel::validate(input, &im2col_input, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false));
+        input_to_use = &im2col_input;
+    }
+    else
+    {
+        // Fully Connected layer after a Fully Connected Layer without batches
+        ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(0) != weights_to_use->dimension(1));
+    }
+    // Validate matrix multiply kernel
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_mm(*input_to_use, *weights_to_use, *tmp_output, false));
+
+    // Validate output stage for asymmetric quantized types
+    if(is_quantized)
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint::validate(&gemmlowp_output, biases, output));
+    }
+
+    return Status{};
+}
+
 void CLFullyConnectedLayer::run()
 {
     // Reshape of the weights (happens only once)
diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
index 5c6f5b4ed0..ddcab6a256 100644
--- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
@@ -29,9 +29,11 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 
 using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
 
 CLGEMMLowpMatrixMultiplyCore::CLGEMMLowpMatrixMultiplyCore(std::shared_ptr<IMemoryManager> memory_manager)
     : _memory_group(std::move(memory_manager)), _mm_kernel(), _mtx_a_reshape_kernel(), _mtx_b_reshape_kernel(), _mtx_a_reduction_kernel(), _mtx_b_reduction_kernel(), _offset_contribution_kernel(),
@@ -41,14 +43,9 @@ CLGEMMLowpMatrixMultiplyCore::CLGEMMLowpMatrixMultiplyCore(std::shared_ptr<IMemo
 
 void CLGEMMLowpMatrixMultiplyCore::configure(const ICLTensor *a, const ICLTensor *b, ICLTensor *output, const GEMMInfo &gemm_info)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::QASYMM8);
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(a, b);
-    ARM_COMPUTE_ERROR_ON_MSG((a)->info()->dimension(0) != (b)->info()->dimension(1), "The product AB is defined only if the number of columns in A is equal to the number of rows in B");
-    ARM_COMPUTE_ERROR_ON_MSG((a)->info()->dimension(1) != (output)->info()->dimension(1), "The output matrix must have the same number of rows as the matrix A");
-    ARM_COMPUTE_ERROR_ON_MSG((b)->info()->dimension(0) != (output)->info()->dimension(0), "The output matrix must have the same number of columns as the matrix B");
-    ARM_COMPUTE_ERROR_ON_MSG(gemm_info.is_a_reshaped(), "Matrix A already reshaped is not supported");
-    ARM_COMPUTE_ERROR_ON_MSG(gemm_info.is_b_reshaped(), "Matrix B already reshaped is not supported");
+    ARM_COMPUTE_ERROR_ON_NULLPTR(a, b, output);
+    ARM_COMPUTE_UNUSED(gemm_info);
+    ARM_COMPUTE_ERROR_THROW_ON(CLGEMMLowpMatrixMultiplyCore::validate(a->info(), b->info(), output->info(), gemm_info));
 
     _reshape_b_only_on_first_run = gemm_info.reshape_b_only_on_first_run();
     _a_offset                    = a->info()->quantization_info().offset;
@@ -65,18 +62,8 @@ void CLGEMMLowpMatrixMultiplyCore::configure(const ICLTensor *a, const ICLTensor
         matrix_a = &_tmp_a;
         matrix_b = &_tmp_b;
 
-        // The interleaved output matrix will have the following shape: [ a_height * 4, ceil(a_width / 4.0f) ]
-        TensorShape shape_tmp_a = a->info()->tensor_shape();
-        shape_tmp_a.set(0, a->info()->dimension(0) * 4);
-        shape_tmp_a.set(1, std::ceil(a->info()->dimension(1) / 4.f));
-
-        // The transpose1xW output matrix will have the following shape: [ b_height * 16, ceil(b_width / 16.0f) ]
-        TensorShape shape_tmp_b = b->info()->tensor_shape();
-        shape_tmp_b.set(0, b->info()->dimension(1) * 16);
-        shape_tmp_b.set(1, std::ceil(b->info()->dimension(0) / 16.f));
-
-        TensorInfo info_a(shape_tmp_a, 1, a->info()->data_type());
-        TensorInfo info_b(shape_tmp_b, 1, b->info()->data_type());
+        TensorInfo info_a(compute_interleaved_shape(*a->info()), 1, a->info()->data_type());
+        TensorInfo info_b(compute_transpose1xW_shape(*b->info()), 1, b->info()->data_type());
         _tmp_a.allocator()->init(info_a);
         _tmp_b.allocator()->init(info_b);
         _memory_group.manage(&_tmp_a);
@@ -95,13 +82,7 @@ void CLGEMMLowpMatrixMultiplyCore::configure(const ICLTensor *a, const ICLTensor
     // Initialize matrix B reduction kernel only if _a_offset is not equal to 0
     if(_a_offset != 0)
     {
-        TensorShape shape_vector_sum_col = b->info()->tensor_shape();
-
-        if(shape_vector_sum_col.num_dimensions() > 1)
-        {
-            shape_vector_sum_col.remove_dimension(1);
-        }
-        TensorInfo info_vector_sum_col(shape_vector_sum_col, 1, DataType::S32);
+        TensorInfo info_vector_sum_col(compute_reductionA_shape(*b->info()), 1, DataType::S32);
         _vector_sum_col.allocator()->init(info_vector_sum_col);
         _memory_group.manage(&_vector_sum_col);
 
@@ -112,13 +93,7 @@ void CLGEMMLowpMatrixMultiplyCore::configure(const ICLTensor *a, const ICLTensor
     // Initialize Matrix A reduction kernel only if _b_offset is not equal to 0
     if(_b_offset != 0)
     {
-        TensorShape shape_vector_sum_row = a->info()->tensor_shape();
-        shape_vector_sum_row.set(Window::DimX, a->info()->dimension(1));
-        if(a->info()->num_dimensions() > 1)
-        {
-            shape_vector_sum_row.remove_dimension(1);
-        }
-        TensorInfo info_vector_sum_row(shape_vector_sum_row, 1, DataType::S32);
+        TensorInfo info_vector_sum_row(compute_reductionB_shape(*a->info()), 1, DataType::S32);
         _vector_sum_row.allocator()->init(info_vector_sum_row);
         _memory_group.manage(&_vector_sum_row);
 
@@ -147,6 +122,67 @@ void CLGEMMLowpMatrixMultiplyCore::configure(const ICLTensor *a, const ICLTensor
     }
 }
 
+Status CLGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *output, const GEMMInfo &gemm_info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::QASYMM8);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(a, b);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG((a)->dimension(0) != (b)->dimension(1),
+                                    "The product AB is defined only if the number of columns in A is equal to the number of rows in B");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG((a)->dimension(1) != (output)->dimension(1),
+                                    "The output matrix must have the same number of rows as the matrix A");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG((b)->dimension(0) != (output)->dimension(0),
+                                    "The output matrix must have the same number of columns as the matrix B");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_a_reshaped(), "Matrix A already reshaped is not supported");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_b_reshaped(), "Matrix B already reshaped is not supported");
+
+    int32_t a_offset                  = a->quantization_info().offset;
+    int32_t b_offset                  = b->quantization_info().offset;
+    bool    is_interleaved_transposed = a->dimension(1) > 16;
+
+    if(is_interleaved_transposed)
+    {
+        TensorInfo info_a(compute_interleaved_shape(*a), 1, a->data_type());
+        TensorInfo info_b(compute_transpose1xW_shape(*b), 1, b->data_type());
+
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMInterleave4x4Kernel::validate(a, &info_a));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMTranspose1xWKernel::validate(b, &info_b));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyKernel::validate(&info_a, &info_b, output));
+    }
+    else
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyKernel::validate(a, b, output));
+    }
+
+    TensorInfo info_vector_sum_col, info_vector_sum_row;
+
+    // Validate matrix B reduction kernel only if _a_offset is not equal to 0
+    if(a_offset != 0)
+    {
+        info_vector_sum_col = TensorInfo(compute_reductionA_shape(*b), 1, DataType::S32);
+
+        // Configure Matrix B reduction kernel
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixBReductionKernel::validate(b, &info_vector_sum_col));
+    }
+
+    // Validate Matrix A reduction kernel only if _b_offset is not equal to 0
+    if(b_offset != 0)
+    {
+        info_vector_sum_row = TensorInfo(compute_reductionB_shape(*a), 1, DataType::S32);
+
+        // Configure matrix A reduction kernel
+        ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(a, &info_vector_sum_row));
+    }
+
+    // Validate offset contribution kernel
+    ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOffsetContributionKernel::validate(output,
+                                                                             a_offset == 0 ? nullptr : &info_vector_sum_col,
+                                                                             b_offset == 0 ? nullptr : &info_vector_sum_row,
+                                                                             a_offset, b_offset));
+
+    return Status{};
+}
+
 void CLGEMMLowpMatrixMultiplyCore::run()
 {
     _memory_group.acquire();
diff --git a/tests/validation/CL/FullyConnectedLayer.cpp b/tests/validation/CL/FullyConnectedLayer.cpp
index 0d8c8774b8..aba92f14d8 100644
--- a/tests/validation/CL/FullyConnectedLayer.cpp
+++ b/tests/validation/CL/FullyConnectedLayer.cpp
@@ -115,6 +115,57 @@ DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(frame
     ARM_COMPUTE_EXPECT(weights.info()->quantization_info() == weights_quantization_info, framework::LogLevel::ERRORS);
 }
 
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(
+    framework::dataset::make("InputInfo", { TensorInfo(TensorShape(9U, 5U, 7U, 3U), 1, DataType::F32),    // Mismatching data types
+                                            TensorInfo(TensorShape(9U, 5U, 7U, 3U), 1, DataType::QS8, 2), // Mismatching fixed point position
+                                            TensorInfo(TensorShape(8U, 4U, 6U, 4U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(8U, 4U, 6U, 4U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(9U, 5U, 7U, 3U), 1, DataType::F32),    // Invalid weights dimensions
+                                            TensorInfo(TensorShape(9U, 5U, 7U, 3U), 1, DataType::F32),    // Wrongly reshaped weights
+                                            TensorInfo(TensorShape(8U, 4U, 6U, 4U), 1, DataType::F32),
+                                          }),
+    framework::dataset::make("WeightsInfo",{ TensorInfo(TensorShape(315U, 271U), 1, DataType::F16),
+                                             TensorInfo(TensorShape(315U, 271U), 1, DataType::QS8, 3),
+                                             TensorInfo(TensorShape(192U, 192U), 1, DataType::F32),
+                                             TensorInfo(TensorShape(192U, 192U), 1, DataType::F32),
+                                             TensorInfo(TensorShape(217U, 315U), 1, DataType::F32),
+                                             TensorInfo(TensorShape(217U, 315U), 1, DataType::F32),
+                                             TensorInfo(TensorShape(192U, 192U), 1, DataType::F32),
+                                          })),
+    framework::dataset::make("BiasInfo",{ TensorInfo(TensorShape(271U), 1, DataType::F32),
+                                          TensorInfo(TensorShape(271U), 1, DataType::QS8, 2),
+                                          TensorInfo(TensorShape(192U), 1, DataType::F32),
+                                          TensorInfo(TensorShape(192U), 1, DataType::F32),
+                                          TensorInfo(TensorShape(271U), 1, DataType::F32),
+                                          TensorInfo(TensorShape(271U), 1, DataType::F32),
+                                          TensorInfo(TensorShape(192U), 1, DataType::F32),
+                                          })),
+    framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(271U, 3U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(271U, 3U), 1, DataType::QS8, 3),
+                                            TensorInfo(TensorShape(192U, 4U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(192U, 4U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(271U, 3U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(271U, 3U), 1, DataType::F32),
+                                            TensorInfo(TensorShape(192U, 4U), 1, DataType::F32),
+                                           })),
+    framework::dataset::make("TransposeWeights",{ true, true, true, false, true, true, true })),
+    framework::dataset::make("ReshapedWeights",{ false, false, false, false, false, false , false})),
+    framework::dataset::make("Expected", { false, false, true, true, false, false, true })),
+    input_info, weights_info, bias_info, output_info, transpose_weights, reshaped_weights, expected)
+{
+    Status status = CLFullyConnectedLayer::validate(&input_info.clone()->set_is_resizable(false),
+                                                    &weights_info.clone()->set_is_resizable(false),
+                                                    &bias_info.clone()->set_is_resizable(false),
+                                                    &output_info.clone()->set_is_resizable(false),
+                                                    transpose_weights,
+                                                    reshaped_weights);
+    ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
 template <typename T>
 using CLFullyConnectedLayerFixture = FullyConnectedLayerValidationFixture<CLTensor, CLAccessor, CLFullyConnectedLayer, T, false>;
 
diff --git a/tests/validation/NEON/Col2Im.cpp b/tests/validation/NEON/Col2Im.cpp
index 9125dc2498..9f2415d628 100644
--- a/tests/validation/NEON/Col2Im.cpp
+++ b/tests/validation/NEON/Col2Im.cpp
@@ -58,8 +58,8 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
                framework::dataset::make("Expected", { false, false, false, false, true })),
                input_info, output_info, convolved_width, convolved_height, expected)
 {
-    bool err = bool(NECol2Im::validate(&input_info, &output_info, Size2D(convolved_width, convolved_height)));
-    ARM_COMPUTE_EXPECT(err == expected, framework::LogLevel::ERRORS);
+    bool status = bool(NECol2Im::validate(&input_info, &output_info, Size2D(convolved_width, convolved_height)));
+    ARM_COMPUTE_EXPECT(status == expected, framework::LogLevel::ERRORS);
 }
 // clang-format on
 // *INDENT-ON*
diff --git a/tests/validation/NEON/Im2Col.cpp b/tests/validation/NEON/Im2Col.cpp
index 4faa7d7d66..f8e474b6c3 100644
--- a/tests/validation/NEON/Im2Col.cpp
+++ b/tests/validation/NEON/Im2Col.cpp
@@ -56,8 +56,8 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
                framework::dataset::make("Expected", { false, false, false, false, true })),
                input_info, output_info, has_bias, expected)
 {
-    bool err = bool(NEIm2Col::validate(&input_info, &output_info, Size2D(3U, 3U), PadStrideInfo(), has_bias));
-    ARM_COMPUTE_EXPECT(err == expected, framework::LogLevel::ERRORS);
+    bool status = bool(NEIm2Col::validate(&input_info, &output_info, Size2D(3U, 3U), PadStrideInfo(), has_bias));
+    ARM_COMPUTE_EXPECT(status == expected, framework::LogLevel::ERRORS);
 }
 // clang-format on
 // *INDENT-ON*