TinyML is a branch of machine learning focused on developing models that run efficiently on low-power, resource-constrained devices like microcontrollers. The recent progress in TinyML technologies triggers the need to address the challenge of balancing inference time and classification quality. These constraints emphasize the need for specialized optimization techniques when implementing Machine Learning (ML) applications on such platforms. While deep neural networks are widely used in TinyML, Dynamic Ensemble Selection (DES) methods offer promising alternatives for improving adaptability and performance. This study examined a proposal for dealing with traffic-varying data streams in systems with constrained resources. Experiments indicate that our method enables the processing of a larger number of images under varying traffic conditions and concept drift, albeit with slightly reduced accuracy.