M2DA: A Low-Complex Design Methodology for Convolutional Neural Network Exploiting Data Symmetry and Redundancy