This SSRN preprint proposes MLP architectures hardwired to compute Lp norms and distances directly, providing a hardware-friendly alternative to learned distance representations. The approach is designed for efficient edge inference where distance-based computations—such as nearest-neighbour search—are commonly required.