COMPMID-617: Add validate support for NEON FullyConnectedLayer

Change-Id: I08987022c8d4cc335c00b8af27bd3edb8fe64d3b
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111596
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Alexander Gilday <alexander.gilday@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

21 files changed, 535 insertions, 202 deletions

diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h
index 051b8b7d4f..e48a9a77e4 100644
--- a/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h
+++ b/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h
@@ -55,6 +55,14 @@ public:
      * @param[in]      biases The shared biases tensor to append. It must be 1D Tensor. Data type supported: Same as @p input
      */
     void configure(ITensor *accum, const ITensor *biases);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMMatrixAccumulateBiasesKernel
+     *
+     * @param[in] accum  The accumulate tensor to convert. Data type supported: QS8/QS16/F32
+     * @param[in] biases The shared biases tensor to append. It must be 1D Tensor. Data type supported: Same as @p input
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *accum, const ITensorInfo *biases);
 
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h
index 4598e15b8e..d54522c678 100644
--- a/arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h
+++ b/arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h
@@ -58,22 +58,28 @@ public:
      * @note If the output tensor is a matrix, the input matrices @p input0 and @p input1 should be the output of the kernels: @ref NEGEMMInterleave4x4Kernel and @ref NEGEMMTranspose1xWKernel
      *       These two kernels change the layout of the original matrices to be more cache-friendly.
      *
-     * @param[in]  input0 Input tensor containing the interleaved Matrix A or the vector A. Data types supported: QS8/QS16/F16/F32
-     * @param[in]  input1 Input tensor containing the transposed Matrix B if the first input tensor A is not a vector.
-     *                    If the output tensor is a vector, input1 must contain the matrix B not reshaped. Data type supported: same as @p input0
-     * @param[out] 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]  input0         Input tensor containing the interleaved Matrix A or the vector A. Data types supported: QS8/QS16/F16/F32
+     * @param[in]  input1         Input tensor containing the transposed Matrix B if the first input tensor A is not a vector.
+     *                            If the output tensor is a vector, input1 must contain the matrix B not reshaped. Data type supported: same as @p input0
+     * @param[out] 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 (Optional) True if input0 and input1 have been reshaped respectively using @ref NEGEMMInterleave4x4Kernel and @ref NEGEMMTranspose1xWKernel
+     * @param[in]  reshape_info   (Optional) GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped
      */
-    void configure(const ITensor *input0, const ITensor *input1, ITensor *output, float alpha);
+    void configure(const ITensor *input0, const ITensor *input1, ITensor *output, float alpha, bool is_interleaved, const GEMMReshapeInfo &reshape_info = GEMMReshapeInfo());
     /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMMatrixMultiplyKernel
      *
-     * @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] input0         Input tensor containing the interleaved Matrix A or the vector A. Data types supported: QS8/QS16/F16/F32
+     * @param[in] input1         Input tensor containing the transposed Matrix B if the first input tensor A is not a vector.
+     *                           If the output tensor is a vector, input1 must contain the matrix B not reshaped. 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 (Optional) True if input0 and input1 have been reshaped respectively using @ref NEGEMMInterleave4x4Kernel and @ref NEGEMMTranspose1xWKernel
+     * @param[in] reshape_info   (Optional) GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output);
+    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved, const GEMMReshapeInfo &reshape_info);
 
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
diff --git a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h
index 6d11fbee11..1659b725bb 100644
--- a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h
+++ b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h
@@ -76,28 +76,34 @@ 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
-     *                         Note: QASYMM8 works only for has_bias = false
-     * @param[out] output      The output tensor. 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.
+     * @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
+     *                                Note: QASYMM8 works only for has_bias = false
+     * @param[out] output             The output tensor. 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.
+     * @param[in]  is_fully_connected Determines whether this kernel will be called by @ref NEFullyConnectedLayer in order to validate the arguments
+     * @param[in]  is_flatten         (Optional) Determines whether this kernel will be called by @ref NEFlattenLayer in order to validate the arguments
      */
-    void configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias);
+    void configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
+                   bool has_bias, bool is_fully_connected = false, bool is_flatten = false);
     /** Static function to check if given info will lead to a valid configuration of @ref NEIm2ColKernel
      *
-     * @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
-     *                        Note: QASYMM8 works only for has_bias = false
-     * @param[in] output      The output tensor. 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.
+     * @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
+     *                               Note: QASYMM8 works only for has_bias = false
+     * @param[in] output             The output tensor. 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.
+     * @param[in] is_fully_connected Determines whether this kernel will be called by @ref NEFullyConnectedLayer in order to validate the arguments
+     * @param[in] is_flatten         (Optional) Determines whether this kernel will be called by @ref NEFlattenLayer in order to validate the arguments
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
+                           bool has_bias, bool is_fully_connected, bool is_flatten = false);
 
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index c53ac4c71f..e21e5cd0d6 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -157,6 +157,43 @@ inline TensorShape compute_deconvolution_shape(const ITensorInfo &input, unsigne
 
     return scale_out_shape;
 }
+inline TensorShape compute_im2col_shape(const ITensorInfo *input, const int num_input_dimensions = 3)
+{
+    TensorShape output_shape{ input->tensor_shape() };
+
+    output_shape.collapse(num_input_dimensions);
+
+    return output_shape;
+}
+inline TensorShape compute_interleave_custom_shape(const TensorShape &input, const int x_interleave, const int y_interleave)
+{
+    TensorShape output_shape{ input };
+
+    output_shape.set(0, output_shape.x() * x_interleave);
+    output_shape.set(1, std::ceil(output_shape.y() / static_cast<float>(y_interleave)));
+
+    return output_shape;
+}
+
+inline TensorShape compute_fully_connected_reshaped_weights_shape(const ITensorInfo *input, bool transpose_weights, bool is_batched_fc_layer, const int interleave)
+{
+    TensorShape output_shape{ input->tensor_shape() };
+
+    // Transpose weights if the user hasn't done it
+    if(transpose_weights)
+    {
+        output_shape = compute_transposed_shape(*input);
+    }
+
+    // If the we run multiple batches we need 1xW transpose, too.
+    if(is_batched_fc_layer)
+    {
+        output_shape = compute_transposed_shape(input->clone()->set_tensor_shape(output_shape));
+        output_shape = compute_interleave_custom_shape(output_shape, interleave, interleave);
+    }
+
+    return output_shape;
+}
 } // namespace shape_calculator
 } // namespace misc
 } // namespace arm_compute
diff --git a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
index 463a7d53e3..9bc8d21fc4 100644
--- a/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -57,6 +57,16 @@ public:
      * @param[in]  is_batched_fc_layer True if it is a batched fully connected layer
      */
     void configure(const ITensor *input, ITensor *output, bool transpose_weights, bool is_batched_fc_layer);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLFullyConnectedLayerReshapeWeights
+     *
+     * @param[in] input               Weights tensor info. The weights must be 2 dimensional. Data types supported: QS8/QS16/F32.
+     * @param[in] output              Destination tensor info. Data type supported: Same as @p input.
+     * @param[in] transpose_weights   True if the weights must be transposed. Data types supported: Same as @p weights.
+     * @param[in] is_batched_fc_layer True if it is a batched fully connected layer
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, bool transpose_weights, bool is_batched_fc_layer);
 
     // Inherited methods overridden:
     void run() override;
@@ -94,6 +104,18 @@ public:
      * @param[in]  are_weights_reshaped (Optional) Reshape the weights tensor if false. Defaults to false.
      */
     void configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *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 info. Data type supported: QS8/QS16/F16/F32.
+     * @param[in] weights              Weights tensor info. The weights must be 2 dimensional. Data type supported: Same as @p input
+     * @param[in] biases               Bias tensor info. It can be nullptr. Data type supported:Same as @p input.
+     * @param[in] output               Destination tensor info. 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/NEON/functions/NEGEMMConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
index c3c7f825a9..ac5f4caa78 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
@@ -141,12 +141,14 @@ public:
 private:
     /** Configures the appropriate matrix multiply routine
      *
-     * @param[in]  input   Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
-     * @param[in]  weights Weights tensor. Data type supported: Same as @p input.
-     * @param[out] output  Output tensor. Data types supported: Same as @p input,
-     *                     except for input of QASYMM8 type where output should be of S32 type.
+     * @param[in]  input          Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in]  weights        Weights tensor. Data type supported: Same as @p input.
+     * @param[out] output         Output tensor. Data types supported: Same as @p input,
+     *                            except for input of QASYMM8 type where output should be of S32 type.
+     * @param[in]  is_interleaved (Optional) True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel
+     * @param[in]  reshape_info   (Optional) GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped
      */
-    void configure_mm(const ITensor *input, const ITensor *weights, ITensor *output);
+    void configure_mm(const ITensor *input, const ITensor *weights, ITensor *output, bool is_interleaved, const GEMMReshapeInfo &reshape_info = GEMMReshapeInfo());
     /** Prepare the appropriate assembly optimized kernel
      *
      * @param[in] ci CPU information
@@ -178,7 +180,7 @@ private:
     bool _is_fully_connected_convolution;
     bool _are_weights_reshaped;
     bool _is_quantized;
-    bool _is_interleaved_transposed;
+    bool _is_interleaved;
 };
 }
 #endif /* __ARM_COMPUTE_NECONVOLUTIONGEMMLAYER_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMTranspose1xW.h b/arm_compute/runtime/NEON/functions/NEGEMMTranspose1xW.h
index 447b8c9c70..8b9ad136b4 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMTranspose1xW.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMTranspose1xW.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,6 +42,14 @@ public:
      * @param[out] output Output tensor. Data type supported: same as @p input
      */
     void configure(const ITensor *input, ITensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMTranspose1xW
+     *
+     * @param[in] input  First input tensor. Data type supported: U8/S8/QS8/U16/S16/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);
 };
 }
 #endif /*__ARM_COMPUTE_NEGEMMTRANSPOSE1XW_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEIm2Col.h b/arm_compute/runtime/NEON/functions/NEIm2Col.h
index cb08f5cd09..cf4999b5af 100644
--- a/arm_compute/runtime/NEON/functions/NEIm2Col.h
+++ b/arm_compute/runtime/NEON/functions/NEIm2Col.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -39,28 +39,30 @@ class NEIm2Col : public INESimpleFunction
 public:
     /** Configure the im2col NEON 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
-     *                         Note: QASYMM8 works only for has_bias = false
-     * @param[out] output      The output tensor. 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.
+     * @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
+     *                                Note: QASYMM8 works only for has_bias = false
+     * @param[out] output             The output tensor. 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.
+     * @param[in]  is_fully_connected Determines whether this kernel will be called by @ref NEFullyConnectedLayer in order to validate the arguments
      */
-    void configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias);
+    void configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, bool is_fully_connected = false);
     /** Static function to check if given info will lead to a valid configuration of @ref NEIm2Col
      *
-     * @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
-     *                        Note: QASYMM8 works only for has_bias = false
-     * @param[in] output      The output tensor. 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.
+     * @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
+     *                               Note: QASYMM8 works only for has_bias = false
+     * @param[in] output             The output tensor. 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.
+     * @param[in] is_fully_connected Determines whether this kernel will be called by @ref NEFullyConnectedLayer in order to validate the arguments
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, bool is_fully_connected);
 };
 }
 #endif /* __ARM_COMPUTE_NEIM2COL_H__ */
diff --git a/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
index 3dd59bddd6..cab3c7a58f 100644
--- a/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -39,6 +39,42 @@
 
 using namespace arm_compute;
 
+namespace
+{
+inline 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_POSITION(biases, accum);
+    ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
+    ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != accum->dimension(0));
+
+    return Status{};
+}
+
+inline std::pair<Status, Window> validate_and_configure_window(ITensorInfo *accum, ITensorInfo *biases)
+{
+    constexpr unsigned int num_elems_processed_per_iteration = 16;
+
+    // Configure kernel window
+    Window win = calculate_max_window(*accum, Steps(num_elems_processed_per_iteration));
+
+    bool window_changed = update_window_and_padding(win,
+                                                    AccessWindowHorizontal(accum, 0, num_elems_processed_per_iteration),
+                                                    AccessWindowStatic(biases, 0, 0, ceil_to_multiple(biases->dimension(0), num_elems_processed_per_iteration), biases->tensor_shape().y()));
+
+    AccessWindowHorizontal output_access(accum, 0, num_elems_processed_per_iteration);
+
+    // Set the valid region for the accum tensor
+    Coordinates coord;
+    coord.set_num_dimensions(accum->num_dimensions());
+    output_access.set_valid_region(win, ValidRegion(coord, accum->tensor_shape()));
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
 NEGEMMMatrixAccumulateBiasesKernel::NEGEMMMatrixAccumulateBiasesKernel()
     : _accum(nullptr), _biases(nullptr)
 {
@@ -46,31 +82,26 @@ NEGEMMMatrixAccumulateBiasesKernel::NEGEMMMatrixAccumulateBiasesKernel()
 
 void NEGEMMMatrixAccumulateBiasesKernel::configure(ITensor *accum, const ITensor *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_POSITION(biases, accum);
-    ARM_COMPUTE_ERROR_ON(biases->info()->num_dimensions() > 1);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(accum, biases);
+
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(accum->info(), biases->info()));
 
     _biases = biases;
     _accum  = accum;
 
-    constexpr unsigned int num_elems_processed_per_iteration = 16;
-
     // Configure kernel window
-    Window win = calculate_max_window(*accum->info(), Steps(num_elems_processed_per_iteration));
-
-    update_window_and_padding(win,
-                              AccessWindowHorizontal(accum->info(), 0, num_elems_processed_per_iteration),
-                              AccessWindowStatic(biases->info(), 0, 0, ceil_to_multiple(biases->info()->dimension(0), num_elems_processed_per_iteration), biases->info()->tensor_shape().y()));
-
-    AccessWindowHorizontal output_access(accum->info(), 0, num_elems_processed_per_iteration);
+    auto win_config = validate_and_configure_window(accum->info(), biases->info());
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    INEKernel::configure(win_config.second);
+}
 
-    // Set the valid region for the accum tensor
-    Coordinates coord;
-    coord.set_num_dimensions(accum->info()->num_dimensions());
-    output_access.set_valid_region(win, ValidRegion(coord, accum->info()->tensor_shape()));
+Status NEGEMMMatrixAccumulateBiasesKernel::validate(const ITensorInfo *accum, const ITensorInfo *biases)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(accum, biases));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(accum->clone().get(), biases->clone().get()).first);
 
-    INEKernel::configure(win);
+    return Status{};
 }
 
 void NEGEMMMatrixAccumulateBiasesKernel::run(const Window &window, const ThreadInfo &info)
diff --git a/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp
index aa5e2dd0dd..69b052a9bd 100644
--- a/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -36,6 +36,8 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
 
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
 #include <arm_neon.h>
 #include <cstddef>
 #include <cstdint>
@@ -1409,27 +1411,73 @@ void matrix_matrix_multiply_qs16(const ITensor *input0, const ITensor *input1, I
     ina, inb, out);
 }
 
-Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
+inline Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved, const GEMMReshapeInfo &reshape_info)
 {
+    ARM_COMPUTE_UNUSED(alpha);
+
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::F16, DataType::F32, DataType::QS8, DataType::QS16);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1, output);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input0, input1, output);
-    ARM_COMPUTE_UNUSED(input0);
-    ARM_COMPUTE_UNUSED(input1);
-    ARM_COMPUTE_UNUSED(output);
 
-    if(output->dimension(1) == 1)
+    if(!is_interleaved)
     {
         ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(0) != input1->dimension(1));
+
+        if(output->total_size() != 0)
+        {
+            ARM_COMPUTE_RETURN_ERROR_ON(input1->dimension(0) != output->dimension(0));
+            ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(1) != output->dimension(1));
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, output);
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input0, output);
+        }
+    }
+    else
+    {
+        const int m                         = reshape_info.m();
+        const int n                         = reshape_info.n();
+        const int k                         = reshape_info.k();
+        const int mult_transpose1xW_width   = reshape_info.mult_transpose1xW_width();
+        const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height();
+
+        /* Interleave */
+        TensorShape tensor_shape0{ input0->tensor_shape() };
+        tensor_shape0.set(0, k);
+        tensor_shape0.set(1, m);
+
+        const TensorInfo tensor_info0          = input0->clone()->set_tensor_shape(tensor_shape0);
+        const TensorInfo tensor_info_reshaped0 = input0->clone()->set_tensor_shape(misc::shape_calculator::compute_interleaved_shape(tensor_info0, mult_interleave4x4_height));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input0, &tensor_info_reshaped0);
+
+        if(n != 0) /* Transpose */
+        {
+            TensorShape tensor_shape1{ input1->tensor_shape() };
+            tensor_shape1.set(0, n);
+            tensor_shape1.set(1, k);
+
+            const TensorInfo tensor_info1          = input1->clone()->set_tensor_shape(tensor_shape1);
+            const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(misc::shape_calculator::compute_transpose1xW_with_element_size_shape(tensor_info1, mult_transpose1xW_width));
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input1, &tensor_info_reshaped1);
+        }
+
+        if(output->total_size() != 0)
+        {
+            if(n != 0)
+            {
+                ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(0) != static_cast<size_t>(n));
+            }
+            ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(1) != static_cast<size_t>(m));
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, output);
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input0, output);
+        }
     }
 
     return Status{};
 }
 
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output)
+inline std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output)
 {
-    Window win            = Window();
-    bool   window_changed = false;
+    bool   window_changed{};
+    Window win{};
 
     unsigned int       num_elems_processed_per_iteration_x = 0;
     const unsigned int num_elems_processed_per_iteration_y = 4;
@@ -1538,11 +1586,19 @@ NEGEMMMatrixMultiplyKernel::NEGEMMMatrixMultiplyKernel()
 {
 }
 
-void NEGEMMMatrixMultiplyKernel::configure(const ITensor *input0, const ITensor *input1, ITensor *output, float alpha)
+void NEGEMMMatrixMultiplyKernel::configure(const ITensor *input0, const ITensor *input1, ITensor *output, float alpha, bool is_interleaved, const GEMMReshapeInfo &reshape_info)
 {
-    // Perform validate step
     ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info()));
+
+    // Output tensor auto inizialitation if not yet initialized
+    TensorShape tensor_shape{ input0->info()->tensor_shape() };
+    tensor_shape.set(0, is_interleaved ? reshape_info.n() : input1->info()->dimension(0));
+    tensor_shape.set(1, is_interleaved ? reshape_info.m() : input0->info()->dimension(1));
+
+    auto_init_if_empty(*output->info(), input0->info()->clone()->set_tensor_shape(tensor_shape));
+
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info(), alpha, is_interleaved, reshape_info));
 
     _input0 = input0;
     _input1 = input1;
@@ -1555,9 +1611,10 @@ void NEGEMMMatrixMultiplyKernel::configure(const ITensor *input0, const ITensor
     INEKernel::configure(win_config.second);
 }
 
-Status NEGEMMMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
+Status NEGEMMMatrixMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved,
+                                            const GEMMReshapeInfo &reshape_info)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output, alpha, is_interleaved, reshape_info));
     ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input0->clone().get(), input1->clone().get(), output->clone().get()).first);
 
     return Status{};
diff --git a/src/core/NEON/kernels/NEIm2ColKernel.cpp b/src/core/NEON/kernels/NEIm2ColKernel.cpp
index 633f78de4b..4fa329bf44 100644
--- a/src/core/NEON/kernels/NEIm2ColKernel.cpp
+++ b/src/core/NEON/kernels/NEIm2ColKernel.cpp
@@ -32,6 +32,8 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Validate.h"
 
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
 #include <arm_neon.h>
 #include <cstddef>
 #include <cstdint>
@@ -42,14 +44,34 @@ using namespace arm_compute;
 
 namespace
 {
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
+                          bool has_bias, bool is_fully_connected, bool is_flatten)
 {
     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, output);
     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output);
     ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::QASYMM8 && has_bias);
-    ARM_COMPUTE_UNUSED(kernel_dims);
-    ARM_COMPUTE_UNUSED(conv_info);
+
+    if(is_flatten) /* Called by FlattenLayer */
+    {
+        size_t flatten_shape = input->tensor_shape().x() * input->tensor_shape().y() * input->tensor_shape().z();
+        ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(0) != flatten_shape);
+    }
+    else if(!is_fully_connected) /* Called by ConvolutionLayer */
+    {
+        std::pair<unsigned int, unsigned int> out_dims = scaled_dimensions(input->dimension(0), input->dimension(1), kernel_dims.width, kernel_dims.height, conv_info);
+        ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(0) != (input->dimension(2) * kernel_dims.area() + (has_bias ? 1 : 0)));
+        ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(1) != (out_dims.first * out_dims.second));
+        ARM_COMPUTE_RETURN_ERROR_ON(output->dimension(2) != 1);
+    }
+    else /* Called by FullyConnectedLayer */
+    {
+        const int num_batch_dimensions = std::max(0, static_cast<int>(output->tensor_shape().num_dimensions()) - 1);
+        const int num_input_dimensions = input->tensor_shape().num_dimensions() - num_batch_dimensions;
+
+        TensorInfo expected_output = output->clone()->set_tensor_shape(misc::shape_calculator::compute_im2col_shape(input, num_input_dimensions));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(&expected_output, output);
+    }
 
     return Status{};
 }
@@ -291,12 +313,15 @@ NEIm2ColKernel::NEIm2ColKernel()
 {
 }
 
-void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
+                               bool has_bias, bool is_fully_connected, bool is_flatten)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     // Perform validation step
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), kernel_dims, conv_info, has_bias));
+    ARM_COMPUTE_UNUSED(is_fully_connected);
+    ARM_COMPUTE_UNUSED(is_flatten);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), kernel_dims, conv_info, has_bias, is_fully_connected, is_flatten));
 
     _input          = input;
     _output         = output;
@@ -382,9 +407,10 @@ void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size
     IKernel::configure(window);
 }
 
-Status NEIm2ColKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+Status NEIm2ColKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info,
+                                bool has_bias, bool is_fully_connected, bool is_flatten)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, kernel_dims, conv_info, has_bias));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, kernel_dims, conv_info, has_bias, is_fully_connected, is_flatten));
     return Status{};
 }
 
diff --git a/src/graph/nodes/FullyConnectedLayer.cpp b/src/graph/nodes/FullyConnectedLayer.cpp
index 219e0f9a93..3742150d37 100644
--- a/src/graph/nodes/FullyConnectedLayer.cpp
+++ b/src/graph/nodes/FullyConnectedLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
diff --git a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
index e9d14db96e..2b4670b98c 100644
--- a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
@@ -114,7 +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 = compute_im2col_shape(*input->info());
+    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
@@ -243,7 +243,7 @@ Status CLFullyConnectedLayer::validate(const ITensorInfo *input, const ITensorIn
     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 &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));
 
diff --git a/src/runtime/NEON/functions/NEFlattenLayer.cpp b/src/runtime/NEON/functions/NEFlattenLayer.cpp
index 408eff5746..32edf93b63 100644
--- a/src/runtime/NEON/functions/NEFlattenLayer.cpp
+++ b/src/runtime/NEON/functions/NEFlattenLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -32,6 +32,6 @@ using namespace arm_compute;
 void NEFlattenLayer::configure(const ITensor *input, ITensor *output)
 {
     auto k = arm_compute::support::cpp14::make_unique<NEIm2ColKernel>();
-    k->configure(input, output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false);
+    k->configure(input, output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false, false, true);
     _kernel = std::move(k);
 }
\ No newline at end of file
diff --git a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
index fc04e28972..26b7271710 100644
--- a/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NEFullyConnectedLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,15 +23,18 @@
  */
 #include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h"
 
+#include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/Size2D.h"
 #include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/NEON/NEScheduler.h"
 
 #include <algorithm>
 #include <cmath>
 
-namespace arm_compute
-{
+using namespace arm_compute;
+using namespace arm_compute::misc::shape_calculator;
+
 NEFullyConnectedLayerReshapeWeights::NEFullyConnectedLayerReshapeWeights(std::shared_ptr<IMemoryManager> memory_manager)
     : _memory_group(std::move(memory_manager)), _transpose_kernel(), _transpose1xW_kernel(), _transpose_output(), _transpose_weights(false), _is_batched_fc_layer(false)
 {
@@ -39,13 +42,10 @@ NEFullyConnectedLayerReshapeWeights::NEFullyConnectedLayerReshapeWeights(std::sh
 
 void NEFullyConnectedLayerReshapeWeights::configure(const ITensor *input, ITensor *output, bool transpose_weights, bool is_batched_fc_layer)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON(input->info()->num_dimensions() > 2);
-    ARM_COMPUTE_ERROR_ON(output == nullptr);
-    ARM_COMPUTE_ERROR_ON(!transpose_weights && !is_batched_fc_layer);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
-    const DataType data_type            = input->info()->data_type();
-    const int      fixed_point_position = input->info()->fixed_point_position();
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(NEFullyConnectedLayerReshapeWeights::validate(input->info(), output->info(), transpose_weights, is_batched_fc_layer));
 
     _transpose_weights   = transpose_weights;
     _is_batched_fc_layer = is_batched_fc_layer;
@@ -56,8 +56,7 @@ void NEFullyConnectedLayerReshapeWeights::configure(const ITensor *input, ITenso
         if(_is_batched_fc_layer)
         {
             // Initialize the output tensor for transpose
-            TensorShape shape_transposed(input->info()->dimension(1), input->info()->dimension(0));
-            _transpose_output.allocator()->init(TensorInfo(shape_transposed, 1, data_type, fixed_point_position));
+            _transpose_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_transposed_shape(*input->info())));
             _memory_group.manage(&_transpose_output);
             _transpose_kernel.configure(input, &_transpose_output);
 
@@ -79,11 +78,39 @@ void NEFullyConnectedLayerReshapeWeights::configure(const ITensor *input, ITenso
             // Configure transpose 1xW kernel
             _transpose1xW_kernel.configure(input, output);
         }
+    }
+}
+
+Status NEFullyConnectedLayerReshapeWeights::validate(const ITensorInfo *input, const ITensorInfo *output, bool transpose_weights, bool is_batched_fc_layer)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 2);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(!transpose_weights && !is_batched_fc_layer, "Configuration transpose_weights=false & is_batched_fc_layer=false not supported");
+
+    if(transpose_weights)
+    {
+        if(is_batched_fc_layer)
+        {
+            std::unique_ptr<ITensorInfo> use_output = output->clone();
+            use_output->set_is_resizable(true).reset_padding().set_tensor_shape(compute_transposed_shape(*input));
+
+            ARM_COMPUTE_RETURN_ON_ERROR(NETransposeKernel::validate(input, use_output.get()));
+            ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMTranspose1xWKernel::validate(use_output.get(), output));
+        }
         else
         {
-            ARM_COMPUTE_ERROR("Configuration transpose_weights=false & is_batched_fc_layer=false not supported");
+            ARM_COMPUTE_RETURN_ON_ERROR(NETransposeKernel::validate(input, output));
+        }
+    }
+    else
+    {
+        if(is_batched_fc_layer)
+        {
+            ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMTranspose1xWKernel::validate(input, output));
         }
     }
+
+    return Status{};
 }
 
 void NEFullyConnectedLayerReshapeWeights::run()
@@ -122,26 +149,25 @@ void NEFullyConnectedLayer::configure(const ITensor *input, const ITensor *weigh
     // Weights: flat(In) x Out
     // Biases: Out
     // Output: Out x B (B can be multi-dimensional)
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
 
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, weights, output);
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(NEFullyConnectedLayer::validate(input->info(),
+                                                               weights->info(),
+                                                               biases != nullptr ? biases->info() : nullptr,
+                                                               output->info(),
+                                                               transpose_weights,
+                                                               are_weights_reshaped));
 
-    const DataType data_type            = input->info()->data_type();
-    const int      fixed_point_position = input->info()->fixed_point_position();
-    const int      num_batch_dimensions = std::max(0, static_cast<int>(output->info()->tensor_shape().num_dimensions()) - 1);
-    const int      num_input_dimensions = input->info()->tensor_shape().num_dimensions() - num_batch_dimensions;
-    const size_t   linear_input_size    = input->info()->tensor_shape().total_size_lower(num_input_dimensions);
+    const int    num_batch_dimensions = std::max(0, static_cast<int>(output->info()->tensor_shape().num_dimensions()) - 1);
+    const int    num_input_dimensions = input->info()->tensor_shape().num_dimensions() - num_batch_dimensions;
+    const size_t linear_input_size    = input->info()->tensor_shape().total_size_lower(num_input_dimensions);
 
     _linearize_input      = (input->info()->tensor_shape().x() != linear_input_size) || (num_input_dimensions > 1 && linear_input_size == 1);
     _are_weights_reshaped = are_weights_reshaped;
     _accumulate_biases    = biases != nullptr;
     _is_batched_fc_layer  = num_batch_dimensions > 0;
 
-    // Check if number of batches match
-    ARM_COMPUTE_ERROR_ON(input->info()->tensor_shape().total_size_upper(num_input_dimensions) != output->info()->tensor_shape().total_size_upper(1));
-    ARM_COMPUTE_ERROR_ON(weights->info()->num_dimensions() > 2);
-
     const size_t   interleave_width = 16 / input->info()->element_size();
     const ITensor *weights_to_use   = weights;
 
@@ -149,65 +175,33 @@ void NEFullyConnectedLayer::configure(const ITensor *input, const ITensor *weigh
     {
         weights_to_use = &_reshape_weights_output;
 
-        TensorShape reshaped_weights_shape(weights->info()->tensor_shape());
-
-        // Transpose weights if the user hasn't done it
-        if(transpose_weights)
-        {
-            const size_t shape_x = reshaped_weights_shape.x();
-            reshaped_weights_shape.set(0, reshaped_weights_shape.y());
-            reshaped_weights_shape.set(1, shape_x);
-        }
-
-        // If the we run multiple batches we need 1xW transpose, too.
-        if(_is_batched_fc_layer)
-        {
-            const float shape_x = reshaped_weights_shape.x();
-            reshaped_weights_shape.set(0, reshaped_weights_shape.y() * interleave_width);
-            reshaped_weights_shape.set(1, static_cast<unsigned int>(std::ceil(shape_x / interleave_width)));
-        }
-
-        _reshape_weights_output.allocator()->init(TensorInfo(reshaped_weights_shape, 1, data_type, fixed_point_position));
+        _reshape_weights_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_fully_connected_reshaped_weights_shape(weights->info(),
+                                                  transpose_weights,
+                                                  _is_batched_fc_layer, interleave_width)));
 
         // Reshape the weights
         _reshape_weights_kernel.configure(weights, &_reshape_weights_output, transpose_weights, _is_batched_fc_layer);
     }
 
-    // Check correct shape of weights
-    if(_is_batched_fc_layer)
-    {
-        // Transpose + Transpose1xW
-        ARM_COMPUTE_ERROR_ON(weights_to_use->info()->tensor_shape().x() != linear_input_size * interleave_width);
-        ARM_COMPUTE_ERROR_ON(weights_to_use->info()->tensor_shape().y() != static_cast<unsigned int>(std::ceil(static_cast<float>(output->info()->tensor_shape().x()) / interleave_width)));
-    }
-    else
-    {
-        // Transpose
-        ARM_COMPUTE_ERROR_ON(weights_to_use->info()->tensor_shape().x() != output->info()->tensor_shape().x());
-        ARM_COMPUTE_ERROR_ON(weights_to_use->info()->tensor_shape().y() != linear_input_size);
-    }
-
     const ITensor *multiply_input = input;
 
     if(_linearize_input)
     {
-        TensorShape shape_im2col(input->info()->tensor_shape());
-        shape_im2col.collapse(num_input_dimensions);
-        _im2col_output.allocator()->init(TensorInfo(shape_im2col, 1, data_type, fixed_point_position));
+        _im2col_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_im2col_shape(input->info(), num_input_dimensions)));
 
         // Configure im2col kernel
         _memory_group.manage(&_im2col_output);
-        _im2col_kernel.configure(input, &_im2col_output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false);
+        _im2col_kernel.configure(input, &_im2col_output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false, true);
 
         multiply_input = &_im2col_output;
     }
 
+    int m = multiply_input->info()->dimension(1);
+    int k = multiply_input->info()->dimension(0);
+
     if(_is_batched_fc_layer)
     {
-        TensorShape shape_interleaved(multiply_input->info()->tensor_shape());
-        shape_interleaved.set(0, shape_interleaved.x() * 4);
-        shape_interleaved.set(1, std::ceil(shape_interleaved.y() / 4.f));
-        _interleave4x4_output.allocator()->init(TensorInfo(shape_interleaved, 1, data_type, fixed_point_position));
+        _interleave4x4_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_interleaved_shape(*multiply_input->info())));
 
         // Configure interleave4x4 kernel
         _memory_group.manage(&_interleave4x4_output);
@@ -217,13 +211,10 @@ void NEFullyConnectedLayer::configure(const ITensor *input, const ITensor *weigh
     }
 
     // Configure matrix multiply kernel
-    _mm_kernel.configure(multiply_input, weights_to_use, output, 1.0f);
+    _mm_kernel.configure(multiply_input, weights_to_use, output, 1.0f, _is_batched_fc_layer, GEMMReshapeInfo(m, 0 /* no transpose */, k));
 
     if(_accumulate_biases)
     {
-        ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases);
-        ARM_COMPUTE_ERROR_ON(biases->info()->tensor_shape().x() != output->info()->tensor_shape().x());
-
         // Configure accumulate biases kernel
         _accumulate_biases_kernel.configure(output, biases);
     }
@@ -246,6 +237,88 @@ void NEFullyConnectedLayer::configure(const ITensor *input, const ITensor *weigh
     }
 }
 
+Status NEFullyConnectedLayer::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_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, weights, output);
+
+    const int    num_batch_dimensions = std::max(0, static_cast<int>(output->tensor_shape().num_dimensions()) - 1);
+    const int    num_input_dimensions = input->tensor_shape().num_dimensions() - num_batch_dimensions;
+    const size_t linear_input_size    = input->tensor_shape().total_size_lower(num_input_dimensions);
+
+    const bool linearize_input     = (input->tensor_shape().x() != linear_input_size) || (num_input_dimensions > 1 && linear_input_size == 1);
+    const bool accumulate_biases   = biases != nullptr;
+    const bool is_batched_fc_layer = num_batch_dimensions > 0;
+
+    ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().total_size_upper(num_input_dimensions) != output->tensor_shape().total_size_upper(1));
+    ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 2);
+
+    const size_t                 interleave_width       = 16 / input->element_size();
+    const ITensorInfo           *weights_to_use         = weights;
+    std::unique_ptr<ITensorInfo> reshape_weights_output = input->clone();
+
+    if(!are_weights_reshaped && (transpose_weights || is_batched_fc_layer))
+    {
+        reshape_weights_output->set_tensor_shape(compute_fully_connected_reshaped_weights_shape(weights, transpose_weights, is_batched_fc_layer, interleave_width));
+
+        ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayerReshapeWeights::validate(weights, reshape_weights_output.get(), transpose_weights, is_batched_fc_layer));
+
+        weights_to_use = reshape_weights_output.get();
+    }
+
+    // Check correct shape of weights
+    if(is_batched_fc_layer)
+    {
+        // Transpose + Transpose1xW
+        ARM_COMPUTE_RETURN_ERROR_ON(weights_to_use->tensor_shape().x() != linear_input_size * interleave_width);
+        ARM_COMPUTE_RETURN_ERROR_ON(weights_to_use->tensor_shape().y() != static_cast<unsigned int>(std::ceil(static_cast<float>(output->tensor_shape().x()) / interleave_width)));
+    }
+    else
+    {
+        // Transpose
+        ARM_COMPUTE_RETURN_ERROR_ON(weights_to_use->tensor_shape().x() != output->tensor_shape().x());
+        ARM_COMPUTE_RETURN_ERROR_ON(weights_to_use->tensor_shape().y() != linear_input_size);
+    }
+
+    const ITensorInfo           *multiply_input       = input;
+    std::unique_ptr<ITensorInfo> im2col_output        = input->clone();
+    std::unique_ptr<ITensorInfo> interleave4x4_output = input->clone();
+
+    if(linearize_input)
+    {
+        im2col_output->set_tensor_shape(compute_im2col_shape(input, num_input_dimensions));
+
+        ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, im2col_output.get(), Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false, true));
+
+        multiply_input = im2col_output.get();
+    }
+
+    int m = multiply_input->dimension(1);
+    int k = multiply_input->dimension(0);
+
+    if(is_batched_fc_layer)
+    {
+        interleave4x4_output->set_tensor_shape(compute_interleaved_shape(*multiply_input));
+
+        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMInterleave4x4Kernel::validate(multiply_input, interleave4x4_output.get()));
+
+        multiply_input = interleave4x4_output.get();
+    }
+
+    ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(multiply_input, weights_to_use, output, 1.0f, is_batched_fc_layer, GEMMReshapeInfo(m, 0 /* no transpose */, k)));
+
+    if(accumulate_biases)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases);
+        ARM_COMPUTE_RETURN_ERROR_ON(biases->tensor_shape().x() != output->tensor_shape().x());
+
+        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixAccumulateBiasesKernel::validate(output, biases));
+    }
+
+    return Status{};
+}
+
 void NEFullyConnectedLayer::run()
 {
     // Reshape of the weights (happens only once)
@@ -280,4 +353,3 @@ void NEFullyConnectedLayer::run()
 
     _memory_group.release();
 }
-} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp
index 48a0d2af1c..05907bab07 100644
--- a/src/runtime/NEON/functions/NEGEMM.cpp
+++ b/src/runtime/NEON/functions/NEGEMM.cpp
@@ -120,7 +120,7 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe
 #endif /* defined(__aarch64__) */
         {
             // Configure the matrix multiply kernel
-            _mm_kernel.configure(a, b, d, alpha);
+            _mm_kernel.configure(a, b, d, alpha, false);
         }
 
         // Configure matrix addition kernel
@@ -212,6 +212,10 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe
             _memory_group.manage(&_tmp_a);
             _memory_group.manage(&_tmp_b);
 
+            int m = a->info()->dimension(1);
+            int n = b->info()->dimension(0);
+            int k = a->info()->dimension(0);
+
             // Configure interleave kernel
             _interleave_kernel.configure(a, &_tmp_a);
 
@@ -219,7 +223,7 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe
             _transpose_kernel.configure(b, &_tmp_b);
 
             // Configure matrix multiplication kernel
-            _mm_kernel.configure(&_tmp_a, &_tmp_b, d, alpha);
+            _mm_kernel.configure(&_tmp_a, &_tmp_b, d, alpha, true, GEMMReshapeInfo(m, n, k));
 
             // Allocate once the all configure methods have been called
             _tmp_a.allocator()->allocate();
diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
index d0a16ef40d..a85078cf71 100644
--- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
@@ -178,7 +178,7 @@ TensorShape get_reshaped_weights_shape_conv(const ITensorInfo *weights, bool app
 Status validate_and_initialize_values(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const PadStrideInfo &conv_info, const WeightsInfo &weights_info, DataType &dt,
                                       bool &append_bias,
                                       bool &are_weights_reshaped, unsigned int &kernel_width, unsigned int &kernel_height,
-                                      bool &is_fully_connected_convolution, bool &is_interleaved_transposed, bool &is_quantized,
+                                      bool &is_fully_connected_convolution, bool &is_interleaved, bool &is_quantized,
                                       unsigned int &mat_weights_cols, unsigned int &mat_weights_rows,
                                       unsigned int &conv_w, unsigned int &conv_h)
 {
@@ -219,7 +219,7 @@ Status validate_and_initialize_values(const ITensorInfo *input, const ITensorInf
 
     // Check if its a "fully connected" convolution
     is_fully_connected_convolution = ((conv_w == 1) && (conv_h == 1));
-    is_interleaved_transposed      = (!is_fully_connected_convolution && !is_quantized);
+    is_interleaved                 = (!is_fully_connected_convolution && !is_quantized);
 
     return Status{};
 }
@@ -228,11 +228,11 @@ Status validate_and_initialize_values(const ITensorInfo *input, const ITensorInf
 NEGEMMConvolutionLayer::NEGEMMConvolutionLayer(const std::shared_ptr<IMemoryManager> &memory_manager)
     : _memory_group(memory_manager), _input_im2col_kernel(), _input_interleave_kernel(), _reshape_weights(), _mm_kernel(), _mm_optimised_kernel(nullptr), _mm_gemmlowp(memory_manager),
       _gemmlowp_output_stage(), _output_col2im_kernel(), _input_im2col_reshaped(), _input_interleaved_reshaped(), _weights_reshaped(), _gemm_output(), _tmp_output(), _workspace(), _append_bias(false),
-      _is_fully_connected_convolution(false), _are_weights_reshaped(false), _is_quantized(false), _is_interleaved_transposed(false)
+      _is_fully_connected_convolution(false), _are_weights_reshaped(false), _is_quantized(false), _is_interleaved(false)
 {
 }
 
-void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *weights, ITensor *output)
+void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *weights, ITensor *output, bool is_interleaved, const GEMMReshapeInfo &reshape_info)
 {
     if(_is_quantized)
     {
@@ -252,7 +252,7 @@ void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *w
     }
     else
     {
-        _mm_kernel.configure(input, weights, output, 1.f);
+        _mm_kernel.configure(input, weights, output, 1.f, is_interleaved, reshape_info);
     }
 }
 
@@ -290,7 +290,7 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig
 
     Status status = validate_and_initialize_values(input->info(), weights->info(), (biases == nullptr) ? nullptr : biases->info(), conv_info, weights_info, dt, _append_bias, _are_weights_reshaped,
                                                    kernel_width, kernel_height,
-                                                   _is_fully_connected_convolution, _is_interleaved_transposed, _is_quantized,
+                                                   _is_fully_connected_convolution, _is_interleaved, _is_quantized,
                                                    mat_weights_cols, mat_weights_rows, conv_w, conv_h);
 
     ARM_COMPUTE_ERROR_THROW_ON(status);
@@ -339,9 +339,8 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig
             }
             else
             {
-                const unsigned int transpose_width = 16 / input->info()->element_size();
-                mat_weights_cols                   = weights_info.num_kernels();
-                mat_weights_rows                   = weights->info()->dimension(0) / transpose_width + (_append_bias ? 1 : 0);
+                mat_weights_cols = weights_info.num_kernels();
+                mat_weights_rows = weights_info.kernel_size().first * weights_info.kernel_size().second * input->info()->dimension(2) + (_append_bias ? 1 : 0);
             }
         }
         else
@@ -362,7 +361,7 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig
 
             // Create tensor to store the reshaped weights
             _weights_reshaped.allocator()->init(TensorInfo(reshaped_weights_shape, 1, dt, fixed_point_position));
-            _reshape_weights.configure(weights, biases_to_use, &_weights_reshaped, _is_interleaved_transposed /* 1xW transpose */);
+            _reshape_weights.configure(weights, biases_to_use, &_weights_reshaped, _is_interleaved /* 1xW transpose */);
             weights = &_weights_reshaped;
         }
     }
@@ -430,18 +429,19 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig
     }
     else
     {
-        if(_is_interleaved_transposed)
+        if(_is_interleaved)
         {
             // Configure GEMMInterleave4x4. _input_interleaved_reshaped will be auto configured in the kernel
             _input_interleave_kernel.configure(&_input_im2col_reshaped, &_input_interleaved_reshaped);
 
             // Configure GEMM
-            configure_mm(&_input_interleaved_reshaped, weights, &_gemm_output);
+            configure_mm(&_input_interleaved_reshaped, weights, &_gemm_output, _is_interleaved, GEMMReshapeInfo(_input_im2col_reshaped.info()->dimension(1), 0 /* no transpose */,
+                                                                                                                _input_im2col_reshaped.info()->dimension(0)));
             _input_interleaved_reshaped.allocator()->allocate();
         }
         else
         {
-            configure_mm(&_input_im2col_reshaped, weights, &_gemm_output);
+            configure_mm(&_input_im2col_reshaped, weights, &_gemm_output, _is_interleaved);
         }
     }
 
@@ -479,11 +479,13 @@ void NEGEMMConvolutionLayer::configure(const ITensor *input, const ITensor *weig
 Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
                                         const WeightsInfo &weights_info)
 {
+    ARM_COMPUTE_UNUSED(output);
+
     DataType     dt{};
     bool         append_bias{};
     bool         are_weights_reshaped{};
     bool         is_fully_connected_convolution{};
-    bool         is_interleaved_transposed{};
+    bool         is_interleaved{};
     bool         is_quantized{};
     unsigned int kernel_width     = 0;
     unsigned int kernel_height    = 0;
@@ -493,9 +495,11 @@ Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI
     unsigned int conv_h           = 0;
 
     Status status = validate_and_initialize_values(input, weights, biases, conv_info, weights_info, dt, append_bias, are_weights_reshaped, kernel_width, kernel_height,
-                                                   is_fully_connected_convolution, is_interleaved_transposed, is_quantized, mat_weights_cols, mat_weights_rows,
+                                                   is_fully_connected_convolution, is_interleaved, is_quantized, mat_weights_cols, mat_weights_rows,
                                                    conv_w, conv_h);
 
+    const Size2D kernel_weights = Size2D(kernel_width, kernel_height);
+
     ARM_COMPUTE_RETURN_ON_ERROR(status);
 
     std::unique_ptr<ITensorInfo> reshaped_weights = weights->clone();
@@ -570,7 +574,7 @@ Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI
     shape_im2col.set(1, mat_input_rows);
     shape_im2col.set(2, 1);
     TensorInfo im2_col_info = input->clone()->set_tensor_shape(shape_im2col);
-    ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, &im2_col_info, Size2D(weights->dimension(0), weights->dimension(1)), conv_info, append_bias));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEIm2ColKernel::validate(input, &im2_col_info, kernel_weights, conv_info, append_bias, false));
 
     // Create GEMM output tensor
     TensorShape shape_gemm(im2_col_info.tensor_shape());
@@ -579,24 +583,20 @@ Status NEGEMMConvolutionLayer::validate(const ITensorInfo *input, const ITensorI
     TensorInfo gemm_output_info = input->clone()->set_tensor_shape(shape_gemm);
 
     // Validate GEMM interleave and multiply
-    if(is_interleaved_transposed)
+    if(is_interleaved)
     {
         TensorShape shape_interleaved = shape_im2col;
         shape_interleaved.set(0, shape_interleaved.x() * 4);
         shape_interleaved.set(1, std::ceil(shape_interleaved.y() / 4.f));
         TensorInfo input_interleaved_info = input->clone()->set_tensor_shape(shape_interleaved);
         ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMInterleave4x4Kernel::validate(&im2_col_info, &input_interleaved_info));
-        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(&input_interleaved_info, weights, &gemm_output_info));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(&input_interleaved_info, weights, &gemm_output_info, 1.f, is_interleaved, GEMMReshapeInfo()));
     }
     else
     {
-        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(&im2_col_info, weights, &gemm_output_info));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMMatrixMultiplyKernel::validate(&im2_col_info, weights, &gemm_output_info, 1.f, is_interleaved, GEMMReshapeInfo()));
     }
 
-    ARM_COMPUTE_RETURN_ON_ERROR(NECol2ImKernel::validate(&gemm_output_info, output, Size2D(conv_w, conv_h)));
-
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG((output->dimension(0) != conv_w) || (output->dimension(1) != conv_h), "Output shape does not match the expected one");
-
     return Status{};
 }
 
@@ -621,7 +621,7 @@ void NEGEMMConvolutionLayer::run()
     }
     else
     {
-        if(_is_interleaved_transposed)
+        if(_is_interleaved)
         {
             // Run interleave
             NEScheduler::get().schedule(&_input_interleave_kernel, Window::DimY);
diff --git a/src/runtime/NEON/functions/NEGEMMTranspose1xW.cpp b/src/runtime/NEON/functions/NEGEMMTranspose1xW.cpp
index 571bf2bc74..802b94650e 100644
--- a/src/runtime/NEON/functions/NEGEMMTranspose1xW.cpp
+++ b/src/runtime/NEON/functions/NEGEMMTranspose1xW.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -38,3 +38,7 @@ void NEGEMMTranspose1xW::configure(const ITensor *input, ITensor *output)
     k->configure(input, output);
     _kernel = std::move(k);
 }
+Status NEGEMMTranspose1xW::validate(const ITensorInfo *input, const ITensorInfo *output)
+{
+    return NEGEMMTranspose1xWKernel::validate(input, output);
+}
diff --git a/src/runtime/NEON/functions/NEIm2Col.cpp b/src/runtime/NEON/functions/NEIm2Col.cpp
index 8e90e66dcc..b962db9144 100644
--- a/src/runtime/NEON/functions/NEIm2Col.cpp
+++ b/src/runtime/NEON/functions/NEIm2Col.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,14 +28,14 @@
 
 using namespace arm_compute;
 
-void NEIm2Col::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+void NEIm2Col::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, bool is_fully_connected)
 {
     auto k = arm_compute::support::cpp14::make_unique<NEIm2ColKernel>();
-    k->configure(input, output, kernel_dims, conv_info, has_bias);
+    k->configure(input, output, kernel_dims, conv_info, has_bias, is_fully_connected);
     _kernel = std::move(k);
 }
 
-Status NEIm2Col::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+Status NEIm2Col::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, bool is_fully_connected)
 {
-    return NEIm2ColKernel::validate(input, output, kernel_dims, conv_info, has_bias);
+    return NEIm2ColKernel::validate(input, output, kernel_dims, conv_info, has_bias, is_fully_connected);
 }
diff --git a/tests/validation/NEON/FullyConnectedLayer.cpp b/tests/validation/NEON/FullyConnectedLayer.cpp
index fc5342d304..ed0edcd3be 100644
--- a/tests/validation/NEON/FullyConnectedLayer.cpp
+++ b/tests/validation/NEON/FullyConnectedLayer.cpp
@@ -115,6 +115,52 @@ DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(combine(frame
     validate(dst.info()->valid_region(), dst_valid_region);
 }
 
+// *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 = NEFullyConnectedLayer::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 NEFullyConnectedLayerFixture = FullyConnectedLayerValidationFixture<Tensor, Accessor, NEFullyConnectedLayer, T, true>;
 
diff --git a/tests/validation/NEON/Im2Col.cpp b/tests/validation/NEON/Im2Col.cpp
index f8e474b6c3..96dd6f86ab 100644
--- a/tests/validation/NEON/Im2Col.cpp
+++ b/tests/validation/NEON/Im2Col.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -44,6 +44,7 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
                                                        TensorInfo(TensorShape(10U, 12U, 2U), 1, DataType::F32),     // Mismatching data type
                                                        TensorInfo(TensorShape(10U, 12U, 2U), 1, DataType::QS8, 2),  // Mismatching fixed point
                                                        TensorInfo(TensorShape(10U, 12U, 2U), 1, DataType::QASYMM8), // Bias not supported with QASYMM8
+                                                       TensorInfo(TensorShape(10U, 12U, 2U), 1, DataType::QASYMM8), // Mismatching shapes
                                                        TensorInfo(TensorShape(10U, 12U, 2U), 1, DataType::QASYMM8),
                                                      }),
                framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(3U, 4U, 10U, 2U), 1, DataType::F16),
@@ -51,12 +52,13 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
                                                        TensorInfo(TensorShape(3U, 4U, 10U, 2U), 1, DataType::QS8, 3),
                                                        TensorInfo(TensorShape(3U, 3U, 10U, 2U), 1, DataType::QASYMM8),
                                                        TensorInfo(TensorShape(3U, 4U, 10U, 2U), 1, DataType::QASYMM8),
+                                                       TensorInfo(TensorShape(18U, 80U, 1U, 2U), 1, DataType::QASYMM8),
                                                      })),
-               framework::dataset::make("HasBias", { true, true, true, true, false })),
-               framework::dataset::make("Expected", { false, false, false, false, true })),
+               framework::dataset::make("HasBias", { true, true, true, true, false, false })),
+               framework::dataset::make("Expected", { false, false, false, false, false, true })),
                input_info, output_info, has_bias, expected)
 {
-    bool status = bool(NEIm2Col::validate(&input_info, &output_info, Size2D(3U, 3U), PadStrideInfo(), has_bias));
+    bool status = bool(NEIm2Col::validate(&input_info, &output_info, Size2D(3U, 3U), PadStrideInfo(), has_bias, false));
     ARM_COMPUTE_EXPECT(status == expected, framework::LogLevel::ERRORS);
 }
 // clang-format on