One way you can do this with an RDBMS is to let the relational schema merely be a meta-schema. For example, represent RDF as triples, and have one relational table with three columns, and put everything in there.

There are, of course, disadvantages to this approach. It means that what you think of as the schema is not actually represented by the RDBMS. That means you can’t easily take advantage of the RDBMS’s integrity features, and it’s harder to control physical layout, just to mention two.

You can actually modify the schema of an RDBMS dynamically with DDL statements, but it’s not the “sweet spot”. I have to admit that doing this in ObjectStore (the object-oriented DBMS that I co-designed), which we called “schema evolution”, was also not a piece of cake; it wasn’t designed for that. However, the XML database that we built on top of ObjectStore did provide that kind of flexibility.