This is a larger question than it looks, but you should experiment, bearing in mind you’ll most likely change it as you learn.
Technically, you can use the VSTS agent as your DSC Pull agent instead of the LCM (some have done this with Octopus Deploy for instance, search a DevOpsGuys article by Dave Wyatt iirc), and code whatever logic you need in there. So you can make it more flexible, that’s cool…
Should you do it now?
If it’s the quickest for you to improve your current situation, maybe. Get a minimum viable product out, and learn from it. Understand that it will have to evolve (or it might become tech debt).
Would I do it?
Probably not. Where would you define those groups? How could you re-provision them from code? How can you make them visible, easy to change, enforced? Would those group definition live with the rest of the information? How would you apply the components one by one? Still as independent MOFs?
There’s many challenges to be solved that the LCM and pull model have already (defining intent, eventual convergence, policy-driven, immutable artefacts…), and best practices around Config Data management made easier using Datum…
So it’s one of those: “it depends”… Context is everything.