Yes, but most "architecture engineering" has already started to be phased out in deep learning research, with the exception of specifically tuned network types (i.e. for computational efficiency) or those with special layer types (i.e. separable convolutions, deconvolutions, etc.).