Distinguishing small and large allocations has critical impact on the long term performance of TinyML deployments.