Actually, I realized that you were more than correct.
In fact, I produced a full PL/SQL parser (based on SQL-2003) and the size in-memory for the irony grammar was huge (25-30Mb), sending all the parameters off-load. Also, the execution slowed down sensibly and I realized this is not the right decision.
I started writing a new parser based on ANTLR4, that appears to be very performant and small on memory (less than 5Mb) still too early for a final response, as I've written just the Expression parsing (that is, anyway, the most difficult part).
What it worries me the most is the final size of the library: the expression parser is so far 378Kb, while the ANTLR4 runtime library is little more than 200kb: I foresee that the final SQL parser library will reach 800kb and that will bring the only SQL parsing domain up to 1Mb.
My intention is to embed everything through merging into the kernel library (deveeldb.dll) that would add the 1Mb or remaining code (also other 2 small libraries embedded: one for IoC and another for Astronomical Mathematics) to the SQL. That makes it 2Mb... not nice in an embedded context (despite of the performances).
I'm considering an alternative version (to name "micro") that doesn't include the SQL parsing (and other features, like security), to produce a version under 1Mb.