What are the key considerations when designing deep learning architectures for large-scale image processing tasks?