There's no single tool that will do this for you, as far as I know. You generally need additional information (multispectral imagery, terrestrial ecosystem/forestry mapping, etc.) to build a model, incorporating things you know about what you want to model (height, density, smoothness, reflective characteristics, etc.).
Not sure the exact steps I would take, but you should look at taking your building footprints and running the Extracting Buildings and Trees from Lidar tool (BuildingTreeIdentification_V3.zip) on my blog along with Trees as points (Trees From LIDAR and NAIP_Pro.TBX.zip). Next, I would use the classification tools with imagery. Segmentation tool seems to be a good way of classifying the result. I would add those as a field to the points and then try to do a group analysis using slope, segmentation, height, width, and maybe NDVI range. I tried this several years ago with some success. Here's an area I segmented:
And then the tree points with attributes run through a group analysis. The Red Triangles were points that are power lines or over water. You can see that different trees species are grouping with the different colors below. You really need ground truth data to do this properly.
You have my number/email if you want to contact me. I have not had time to further this research lately.
Once you have the points with the groups of trees identified, it would be simple to buffer the trees to a polygon and then using the polygons to classify the points based on height.
Arthur Crawford - Esri
A quick "heads up" in case this is a new type of analysis for you - set aside time beforehand to determine what level of accuracy will be acceptable given your project needs and then during the analysis definitely ground truth your results. It is not uncommon with this type of analysis to mis-classify large amounts of data. Expect to do several iterations as the process gets refined. I'd start with a small plot where you already know what trees exists there and see what the analysis says it is, then tweak the settings/process to improve the results.
Also, there always will be some level of error. The question that will need to be answered is what level of error is acceptable.
Chris Donohue, GISP