As mentioned, the key here is the difference between the data and the control - since controls and indicators can also double as function parameters and return values, this might be a bit confusing.
First, it is important to understand that a wire passes only data of its type - it has no meta data which relates to the control it came from (that's not actually correct, but we won't go into that), so you can wire any two controls and indicators as long as their data types are compatible.
As for your questions:
All front panel objects must be allocated in the vi that owns the front panel and they only exist in that panel.
That's essentially correct, but the only thing you need to do to "allocate" them is to drop them on the front panel. When they have data put into them and you change the values of their properties, they will need more area in memory to hold that, but LabVIEW takes care of that behind the scenes.
Objects that are "passed" as indicators(outputs) of the parent into the control (input) of the subvi(child) are not pointers but cause a memcopy into a new instance of the object.
In LabVIEW, the control itself is an object, but the data that comes out of it is separate from the control itself. LabVIEW tries to schedule the execution of the code so that it will have a single copy of any piece of data in memory, but that's not always possible. A reference is a pointer to an existing object. When you use a reference of a certain control class as an input to a subVI, the reference needs to point to an actual existing control. If it does not, any property or invoke node will return an error telling you that you are trying to work on an object which does not exist.
That was kind of stream of consciousness, so feel free to ask for more specific details. If you want,
this article explains this in some more detail.
___________________
Try to take over the world!
