I can't imagine a language more suited for implementing this than Julia. Custom N-d array types are easy to create and are composable with the rest of the linear algebra ecosystem. Multiple index and index iteration strategies are supported. It's easy to implement optimized code for common operations, falling back to reasonably-efficient generic implementations. There are strong metaprogramming facilities in case custom unrolling behaviour is necessary at compile time.
If someone was interested in exploring an implementation I would strongly suggest giving Julia a try.