Compressing Vision Transformers for Low-Resource Visual Learning