In part it depends on the usage of the model, whether for archviz or ultimate design development. For archviz I could imagine a single element that is a continuous shape of building windows, then the visible structure is added to that. The core of the windows could be manipulated as one element.