Tiny Machine Learning (TinyML) has emerged as a transformative technology that enables the deployment of machine learning models on ultra-low-power microcontrollers and resource-constrained Internet of Things (IoT) devices. By performing data processing and inference directly at the edge, TinyML reduces latency, minimizes bandwidth usage, enhances data privacy, and lowers dependence on cloud computing. These advantages make TinyML an ideal solution for smart healthcare, environmental monitoring, industrial automation, agriculture, wearable electronics, and intelligent home applications. However, implementing machine learning algorithms on devices with limited memory, processing capability, and energy resources remains a significant challenge. This paper presents a comprehensive study of TinyML architectures, optimization techniques, deployment strategies, and real-world applications for low-power IoT devices. Various model compression methods, including quantization, pruning, and knowledge distillation, are analyzed to improve computational efficiency while maintaining acceptable prediction accuracy. The paper also discusses hardware platforms, software frameworks, and energy-efficient inference mechanisms that enable real-time intelligent decision-making on edge devices. Experimental analysis demonstrates that TinyML significantly reduces power consumption and communication overhead while improving response time and system reliability. Furthermore, the integration of TinyML with IoT technologies supports scalable and sustainable intelligent systems suitable for next-generation edge computing environments. The study concludes that TinyML is a promising approach for developing efficient, secure, and autonomous low-power IoT applications.