Modeling data set with different dimensions

I am trying to model a dataset with different input dimensions of data instances (varying from 11 to 50) and fixed multi dimensional responses. I am basically having 11 input features, but 4 of these features have variable number of dimensions depending on my data generating process. When I collect data from different systems, due to different dimensions of these 4 features, I have unequal number of total input dimensions in the input dataset. For eg. the 1st data instance may have 15 input variables, the next may have 25 and so on. Is there any way to model such dataset with varying input dimension accurately?

Note that I do not want to use a single metric for the variable dimensional features (such as using mean, variance etc.). Thanks in advance!