If I were you, I'd look for a space that is more restricted than general-purpose programming but still rich enough to do nontrivial computation. If you can find interesting invariants that are true of all computations in such a space (but not true of all programs in general), you have a hope of leveraging them to create both a powerful declarative language and an efficient implementation for it. The space of all programs in general is too large for this.

This is why SQL and spreadsheets have been such successes as declarative programming. If you find another such model, that could be a big deal.