The increase of big data streams creates critical challenges during real-time data classification applications requiring efficient as well as scalable solutions. This research implements new methodology using the Apache Flink framework, which brings together Condensed Nearest Neighbors (CNN) algorithms with Extreme Learning Machine (ELM) to enhance dynamic classification performance. The implementation of CNN decreases dataset volume alongside its dimensions and maintain necessary information. The ELM approach displays rapid training alongside efficient classification execution, which accelerates system efficiency. This research indicates that combining these methods produces high accuracy rates together with minimized computation costs, which makes them ideal for finance and healthcare and sensor-based applications. The proposed method undergoes an extensive evaluation using synthetic big data stream datasets, which shows its successful operation for fast data stream timing durations alongside large data volumes.