Efficient Deep Learning Accelerator Architectures By Model Compression And Data Orchestration