A removal function library is a collection of removal functions with a large number of different shapes, and its establishment plays a crucial role in the invocation of removal functions during computer numerical control polishing. This study focuses on the removal functions of flexible spherical polishing heads, establishing a fitting model based on a super-Gaussian distribution. The model was applied to a least squares fitting of dozens of datasets. Multivariate interpolation methods were used for handling fitting parameters, leading to the creation of a database of removal functions corresponding to any combination of precession angle, pressure, and rotation speed. Four samples were randomly selected from the removal function library and compared with experimental data. The relative error was less than 7% in all cases, confirming the effectiveness of the method.