Big structures, large assemblies for example, usually implies lots of elements even when sensible model 'efficienies' have been made. Then this means longer run times and more resources; particulalry if there are contacts.
One may only be interested in a portion of the overall model and the method usually employed is to refine mesh in only that location but this is not always appropriate; particularly if there are many regions of interest. A better method would be to define the region(s) of interest as a 'submodel(s)'
The 'unrefined' parent model provides the correct stiffenss and therefore force systems into regions of interest and therefore deflections of the regions' boundaries.
Each region is then effectively opened and refined separately and run using an enforced displacement boundary.
This is not new and other software permit this.