Theoretical Comparative Analysis of TinyML Model Architectures and Deployment Frameworks for Resource-Constrained Embedded Systems

Tiny Machine Learning (TinyML) enables the deployment of machine-learning models on low-power microcontrollers. Due to strict constraints on memory, computation, and energy, efficient deployment requires specialized frameworks and lightweight model architectures. This paper presents a theoretical comparison of major TinyML model families—MobileNetV2, SqueezeNet, DS-CNN, Tiny-YOLO, EfficientNet-Lite—and deployment frameworks such as TensorFlow Lite Micro, Edge Impulse, CMSIS-NN, MicroTVM, and uTensor. The study focuses entirely on conceptual principles, architectural design, and theoretical trade-offs, without relying on experimental or empirical evaluation. The objective is to guide researchers and developers in selecting the most suitable TinyML components for resource-constrained embedded systems.