How to make an accurate evaluation of model performance when I only have a small dataset?

Question

The data sizes of my several datasets are only about 100 samples and 10 samples. So if a model misclassificate one sample, it will lead the accur