This is fuzzy because any argument you could make about a particular feature you could make about a whole product. With rare exceptions, Profit and Loss for one product is at least influenced by the success or failure of other products and the company as a whole.
Working with correlation is a large part of understanding P&L. The best way I could recommend looking at it is like bundled products. Let's take office. At some point, Microsoft has to look at P&L for Word, Excel, etc. But they sell it all together as Office. So all sales of the product (or most at least) are actually buying all of them. That doesn't mean they can't separate them. They can look at usage and surveys to get at least a rough idea of which of how much each application is contributing to driving sales. Further, major changes to one product or another can lead to changes in user behavior which can be tracked and correlated to P&L.
So applied to feature, the same ideas work. Users are buying a package of features, but they may buy it at different times. When a feature is released, do you a change in signups? account deletions? Do a lot of users start using that feature - if so, there's customer retention value there. The big problem with this that I encounter in most companies is not that you can't measure this or even that it is difficult. The problem I usually see if that people simply don't. I've worked with dozens of companies and can thing of 2 that actually measure P&L with any rigor or use any kind of modeling in planning. The fact is, a massive percentage of companies are driven completely on gut feeling and opinions.