This work introduces the novel concept of {\em kind refinement},
which we develop in the context of an explicitly
polymorphic ML-like language with type-level computation.
Just as type
refinements embed rich specifications by means of comprehension
principles expressed by predicates over values in the type domain,
kind refinements provide rich {\em kind} specifications by means of
predicates over {\em types} in the kind domain.
By leveraging our powerful refinement kind discipline, types in our
language are not just used to statically classify program
expressions and values, but also conveniently manipulated as
tree-like data structures, with their kinds refined by logical
constraints on such structures. Remarkably, the resulting typing
and kinding disciplines allow for powerful forms of type reflection,
ad-hoc polymorphism and type-directed meta-programming, which are
often found in modern software development, but not typically expressible in a
type-safe manner in general purpose languages.
We validate our approach both formally and pragmatically
by establishing the standard meta-theoretical results of type safety and
via a prototype implementation of a kind checker, type checker and
interpreter for our language.