I will not describe the XSLT-based approach, because I don’t know a single good thing because of the discovery that I see in my own hesitation: "something like (XSLT, maybe?)".
What you need, I tried this in 2004; Ive abandoned XSLT as the language to use when working with XML schema refactoring, because I am of the opinion that the schema object model APIs (SOM / XSOM), APIs designed specifically for analyzing and creating XSD, are much more productive of my time (for an XSLT reader reading this, Ive qualified my expression).
You always have the possibility that what you want is actually simple, therefore, you can use Roll-Your-Own-XSLT or otherwise (economically or intellectually). Exchanging more specific parameters for your problem can help give better answers. However, if you think about it professionally, that is, with minimal assumptions about what your input and output XSDs will look like, you are working for a little work.
What I ended up with was developing a Domain Specific Language for XML Schema refactoring. It is available as part of the QTAssistant XSR Module. Graphically, this is what your “733 line layout of 16 separate files or subcircuits” might look like in QTAssistant.

Its refactoring features allow a lot of what you described from one set of sources. All you need to do is configure the “views” you want with any necessary functions (for example, what happens in an XSD file, automatically pull out the necessary dependencies, change target namespaces, type refactory names, elements, replace basic type remove unwanted particles, introduce sequence wrappers to change power, etc.). A click of a button gives you guaranteed valid XSD files. It also features a command line interface for integration with automated build systems.
In any case, I would consider an API, for example, in .NET Schema , if only to breathe in how XSLT templates can be developed ...