Variable-performance ML algorithms?

I'm curious if there exists proven-algorithms that can perform with variable performance/accuracy. One can think of an example, where an objective function that depends on a parameter that defines the accuracy vs. speed tradeoff. i.e. At certain instances, the algorithm is required to evaluate rapidly, in which case, the accuracy of the result is affected. However, in certain other instances, the algorithm is required to evaluate with maximum accuracy, in which case, it can afford to take long enough to evaluate the result.

I'm aware of anytime-performance algorithms, but I'm curious if there are any others that I haven't stumbled across. Seems fairly useful in a robotics/computer-vision context.

Thanks in advance!