"What if" thoughts on a class of shape use cases

There seems to be some users shape goals could be (crudely) summarized as
wanting an equivalent of RELAX-NG that can be used to *syntactically* check
an *isolated* document's worth of RDF statements.

There is a concrete syntax for RDF that uses XML.

There are libraries to convert between RDF formats.
There is a RELAX-NG style language for XML (I can't remember the name off
the top of my head :)

Can this very specific class of use cases be taken care of by just
converting the RDF to a canonical form of RDF/XML (e.g such that the number
of entities is minimized, or by requiring that the submitted RDF be in a
canonicalized format), and then using RELAX-NG to specify the syntactic
constraints on the XML?

If that would be sufficient to address a subset of use cases, and the case
users are familiar with XML, and do not need or want any of the features
and complexities of RDF qua RDF, this approach could solve their problems
without new tools or recommendations.

It could also simplify the task of developing more RDF like validation
recommendations since the purely syntactic use cases would not have to be
in scope.