As the amount of data involved in scientific research continues to grow, the need for powerful tools for organizing and analyzing this data grows with it. Despite considerable progress in this area by the database research community, the uptake of database technologies within the scientific community has been slow. Contributing to this limited adoption is a tendency to try to build complex, monolithic, total solution-systems, for a community that can rarely afford the resources to tie their existing infrastructures into such a system. This thesis explores two different directions for creating simpler, smaller, more generalpurpose tools for doing data-processing in a scientific computing environment. Grey-Box Probabilistic Databases are an attempt to create a general purpose tool for efficiently integrating database systems with an organization's existing model-building pipelines. By providing a pay-as-you-go approach to the tradeoff between efficiency and integration effort, users can choose how much of their resources to commit as their needs develop. Dynamic Data Management Systems are a new approach to building data processing systems. Instead of a monolithic data-processing infrastructure that typically includes (and has the performance penalties of supporting) functionality that the user does not require, a Dynamic Data Management System constructs entire data-management systems designed specifically to meet the requirements of the user's application.