It's not necessary to model RA4 paper. You're in digital space. All digital color correction is designed to work in linear light, and Lightroom internally is floating point linear light. The more correct thing to do is digitize it into a positive image of the negative in raw samples, apply gain/multipliers to each channel until the film base plus fog is the same exposure, which will render it as light grey to white, then adjust the gamma of each channel so that it's response is linear relative to the exposure that made the density, then invert that. From there, if you did that correctly, a series of grey cards shot over a range of exposures from -5 EV to +7 EV in full stop increments should render grey at each point with little to no color shifts going on. If you got the linearization of each channel right, you should also be able to adjust the exposure in full stop increments in LR and have each respective exposure land at ~46.6% in the Develop module histogram (46.6% is a correctly exposed grey exposure card in LR if you didn't know). If you got that far, you're over 75% of the way there. From there, shoot a MacBeth color checker chart with a ~5500K light (studio strobes are good for this sort of thing as they're as high a CRI as you'll probably get and very close to 5500K, and very consistent), digitize it and apply the previous steps. Up until now, you should be doing all that in raw unmanaged color. At this stage you conform it to a color space, so pick a color space and look up the XYZ to color space matrix for it, most common color spaces are public and readily available. You have to put that matrix into the DNG color matrix metadata so LR can take your samples and convert them to XYZ and render a correct output. From here, you apply a series of per hue angle twists and per hue angle saturation adjustments so that when LR applies the color matrix your colors end up where they're supposed to. If you picked your color space well, you'll have to apply a couple of twists and adjust the saturation, but otherwise, it'll be minimal touching. You can accomplish the same thing with a 3D LUT, but in all honesty, that's a lot of work (and math), and it's a lot simpler to just individually apply the twists with a simple look up table and apply the saturation changes with a different simple lookup table. Plus, if you see something that doesn't look right, adjusting it via a simple little text based lut file is very simple and straight forward. Most changes are a couple more (or less) degrees twist here or there, a little more or less saturation for a given hue angle, etc. Once it's to your liking, that's your profile. As long as you keep your dev in process, it's pretty static for a given emulsion. You'll also need to put the XYZ color coordinates of 5500K into the DNG file so that LR knows what to do for white balance adjustments if you want to change the white balance in the Dev module. The look I supply to customers very closely matches what you'd get from a digital camera that was ran through Adobe Camera Raw, but in reality, you can make it look any way you want. The sky is the limit here.
There are a couple of film specific situations that what I described above doesn't directly address or deal with, however, in practice, it doesn't do awful things to your image and in my experience has largely turned out to be effectively non-issues.
Where most people go wrong is the first thing they do is pull it into PS and use the levels tool to set black and white points then have no choice but to horse around with it to try to get rid of the color casts because they didn't linearize the channels correctly because PS doesn't give you a good way to do that, and once you're in PS, it's using a color space, which means any changes you make are actually relative to the color space it's using. The same goes for white balance adjustments. It doesn't apply gain multipliers the way you think it does, it does them relative to the Correlated Color Temperature line you see on the CIE XYZ chart relative to the color space it's using. If you're already color conformed and you're just trying to change the white balance for a different CCT, that's fine, but if you're trying to do a raw set of multipliers, it's not great.
Colorperfect does a pretty good job if you put the effort into it, but it's ill suited for a high volume environment. NLP can and does work, but has its own set of challenges because it has to work within the framework that LR gives it, and seeing some of the things it's doing, I don't think the author quite has everything nailed down as good as it could be at this point in terms of the finer details. It's very young software, so he should be able to get it there over time. But again, in it's current form, it's ill suited for a high volume environment.