Tiny Machine Learning (TinyML) represents a rapidly emerging paradigm that brings machine learning inference to ultra-low-power and resource-constrained edge devices such as microcontrollers and embedded sensors. This shift toward on-device intelligence enables real-time processing, reduces dependence on cloud infrastructure, and enhances user privacy—making TinyML a compelling solution for a wide range of applications across healthcare, agriculture, environmental monitoring, and industrial IoT. However, deploying ML models in such constrained environments introduces significant challenges in terms of memory, energy efficiency, and latency. This paper presents a comprehensive survey of the TinyML landscape, covering key model optimization techniques such as pruning, quantization, and knowledge distillation, as well as efficient architectures including MobileNet, SqueezeNet, and Tiny-YOLO. We review prominent training and deployment toolchains, including TensorFlow Lite Micro, CMSIS-NN, TVM, and Edge Impulse, and examine their role in enabling practical implementations. Real-world applications are discussed, showcasing how TinyML systems are already impacting diverse domains. We also identify critical research challenges, such as on-device continual learning, explainability, and secure inference, and outline promising future directions involving neuromorphic hardware, blockchain integration, sustainable AI, and theoretical exploration of lightweight LLMs on microcontrollers.