You are tasked with deploying a trained machine learning model on AWS for real-time inference. To optimize both performance and cost, you are considering using AWS Lambda and Amazon SageMaker. What is the best practice for deploying the model to achieve efficient real-time predictions?