The demand for energy in cloud-native applications has increased considerably in recent years. With the rise of container-based deployments for delivering applications, understanding their power usage patterns is critical to lowering them. Unfortunately, cloud vendors do not provide their clients with power consumption details for individual workloads owing to virtualization-related limits inside the cloud infrastructure. This research paper compares the software and hardware-based tools available in the market to measure power consumption and discusses in detail about Kubernetes Efficient Power Level Exporter (Kepler), which addresses the above issue by estimating power metrics at the container level by using extended Berkeley Packet Filter (eBPF) and machine learning (ML) models. Since data-intensive workloads are power-hungry, we run the ML models on a simulated Graphical Processing Unit (GPU) accelerated Kubernetes (K8s) cluster. The metrics extracted by Kepler are carefully analyzed, and the ML workloads are tuned and optimized to use less energy.