The growing demand for real-time decision making in the field under strict energy constraints is pushing conventional digital pipelines toward their limits. In this context, analog computing has re-emerged as a promising paradigm to process information directly in the physical domain, avoiding costly digitization and data transfer [1,2,3]. In particular, mechanical metastructures (MMs) are engineered systems whose functionality emerges from their architecture rather than their constituent material, offering a powerful platform for performing analog computing and inference tasks directly in the mechanical domain. By exploiting elastic interactions and stress propagation, such systems can encode input–output maps and execute computational operations without digitization. This work introduces a computational framework for the design of MMs tailored to analog computing tasks. The framework considers the modeling and optimization of MMs that perform fundamental mathematical operations and inference tasks, demonstrated through representative benchmark examples. The study further introduces the conceptualization of multi-layer MMs, enabling the implementation of algorithms that require sequences of mathematical operations. The proposed approach highlights the potential of mechanical MMs as ultra-low-power computing substrates, opening new perspectives for in-situ computation and edge intelligence in engineering systems.