imo any SQL that requires regression (I have implemented least squares in TSQL w/ recursive CTEs) is probably a bad use case for SQL in general - projecting, filtering, grouping, and sorting is what SQL engines are great at - repeating row access or referencing the previous row on the next row and then running a function is not going to be fast in most of the databases.

To the point on views - they can seem really useful, but many people dont understand that views dont eliminate "useless code" - most engines are going to evaluate everything in the view, and the nested subviews, and of course, the 10 depth views that were created because they are such a great abstraction.

This is where (to me) other non-database programming languages definitely come in - you can codify codegen or other methods that are locked in, not just a huge string and get a lot of the same value as your views (unless of course, everyone is querying the db directly.) If that's the case, then templating out code is still a good plan, but you can usually accomplish it via the client tooling people are using (most of the DB IDEs have pretty decent snippet support.)