You are working on rebuilding your machine learning (ML) pipeline for structured data using Google Cloud. Currently, you utilize PySpark to perform data transformations at scale. However, these pipelines are becoming inefficient, taking over 12 hours to run. To enhance both development speed and pipeline execution time, you prefer to adopt a serverless tool that supports SQL syntax. Additionally, you have already migrated your raw data into Google Cloud Storage. Considering these requirements and constraints, how should you design the new ML pipeline on Google Cloud to optimize both speed and processing efficiency?