Time series classification (TSC) is crucial for edge computing and IoT, enabling intelligent decision-making in resource-constrained environments. Traditional electrical device classification methods, such as Non-Intrusive Load Monitoring (NILM), rely on handcrafted features and centralized processing, leading to latency, high computational costs, and privacy concerns. This study introduces a TinyML-based deep learning framework for real-time electrical device classification directly on edge devices. By leveraging lightweight neural networks, our approach eliminates the need for manual feature extraction while ensuring energy efficiency, scalability, and low-power execution on microcontrollers. TinyML’s ability to run deep learning models locally minimizes cloud dependency, reduces inference latency, and enhances data privacy. To validate our approach, we constructed a custom time series dataset capturing detailed power consumption patterns. Experimental results indicate that TinyML-powered classification achieves robust, efficient, and real-time performance in edge-based TSC. These findings position TinyML as a scalable and energy-efficient alternative for edge-based TSC, effectively addressing NILM’s limitations while unlocking new possibilities in real-time intelligent computing on embedded systems.