Model Acceleration for Efficient Deep Learning Computing